Remove unreachable backend code and fix two defects it was hiding

[roncodes]

Summary

Removes backend code that no input can reach, and fixes two latent defects found while proving that unreachability.

This came out of the coverage campaign in #277. Pushing server/src line coverage to 99.4%+ meant examining every remaining uncovered line, and a large share turned out not to be untested but unreachable — branches shadowed by a broader arm above them, guards on values a type declaration forbids, and fallbacks sitting after an unconditional assignment. No test can execute those, so they block the --fail-under=100 gate permanently.

Two of them were not harmless leftovers. They are fixed here.

Stacked on #277. Base branch is feature/fleetops-server-coverage-100, so this diff shows only the cleanup. Merge #277 first.

The two commits are split deliberately so the behaviour changes can be reviewed apart from the deletions:

Commit	Contents
20fef6bb	The two real defects, with their test updates
2aa143f3	Unreachable-code removal and four deliberate annotations

Defects fixed

1. Invalid coordinates silently reverse-geocoded Null Island

Internal\v1\GeocoderController::reverse() validated coordinates after converting them:

$coordinates = Utils::getPointFromCoordinates($query);   // : Point — never null

if (!$coordinates instanceof Point) {
    return response()->error('Invalid coordinates provided.');   // unreachable
}

getPointFromCoordinates() is typed : Point and returns new Point(0, 0) for unusable input, so the guard never fired and garbage input was reverse-geocoded at 0, 0 instead of being rejected. Now resolved with the existing strict variant (getPointFromCoordinatesStrict(): ?Point) so the guard works as written.

An existing test asserted the old behaviour — reverseCalls[0] === [0.0, 0.0], i.e. it documented the Null Island lookup. It now asserts the error is returned and that no lookup is attempted.

2. Polygon writes could not bind on update

Casts\Polygon returned the raw geometry from its GeometryInterface arm, while Casts\Point and Casts\MultiPolygon return a SpatialExpression. That is not cosmetic:

• SpatialTrait overrides only performInsert — there is no performUpdate. Inserts survive either shape because performInsert() wraps the attribute itself.
• On updates the cast's return value is bound directly, and BaseBuilder::cleanBindings() only expands a SpatialExpression into the WKT + SRID bindings that ST_GeomFromText(?, ?) requires. A bare Geometry passes through untouched, and it has no __toString, so PDO cannot bind it.

Casts\Polygon is now aligned with the other two.

This is the riskiest hunk in the PR and the one worth the closest look. It is isolated in 20fef6bb and can be reverted alone. Note that an insert-based test cannot validate it — the added RulesAndCastsTest case asserts at the binding level and was verified to fail on the old shape and pass on the new one.

Unreachable code removed

Location	Why it could not execute
Casts/{Polygon,MultiPolygon}	concrete-type arms shadowed by instanceof GeometryInterface (Polygon → MultiLineString → GeometryCollection → Geometry implements GeometryInterface)
Models/Place::createFromMixed	re-tested isCoordinatesStrict() inside a branch already gated on it one arm above
Models/Place::insertFromMixed	instanceof GoogleAddress arm sat below is_array || is_object, which swallows every object — reordered so a GoogleAddress routes to insertFromGoogleAddress() instead of being flattened to an array
Models/Place::insertFromMixed	address-key check sat in the is_string($place) branch, where empty() on a non-numeric string offset is always true — moved into the array branch where an address key can exist
Integrations/Lalamove::__callStatic	instance is a declared public static, so PHP never routes it here
Models/ServiceRate::getLngLatFromPlace	guarded getLocationAsPoint(): SpatialPoint, non-nullable, and a SpatialPoint always exposes getLat/getLng
Internal/v1/MetricsController::resolvePeriod	guarded $request->date(), which returns ?Carbon; both ?? fallbacks yield DateTime
Resources/v1/{Maintenance,MaintenanceSchedule,WorkOrder,Order}	JsonResource fallbacks after Find::httpResourceForModel(), which always resolves a class (falls back to FleetbaseResource)
Resources/v1/{PurchaseRate,TrackingStatus}	middle branch guarded by method_exists($this, 'loadMissing'); no class in the hierarchy declares it — it only resolves via __call, which method_exists cannot see
Api/v1/DriverController::create	company re-check after an early return already guaranteed it
Api/v1/OrderController::create	type fallback after line 72 assigns it unconditionally — replaced with ?? 'transport' at the assignment
Models/Payload::setPlace	createFromMixed(): ?Place always yields a Model, so the Str::isUuid() arm and trailing else were both shadowed
Support/Utils::coordsToCircle	loop ran 0..360 inclusive, so the ring was already closed and the closing push never ran — loop is now exclusive, which also removes a duplicated vertex. Output is otherwise byte-identical (verified: 121 points, closed, zero duplicate interior vertices)

Kept and annotated, not deleted

Four guards are unreachable today but deliberately retained with @codeCoverageIgnore and an explanatory comment:

• Casts/Point's SpatialExpression arm. The guard that shadows it (instanceof Expression) returns without recording $model->geometries[$key], so deleting it would erase evidence of a real asymmetry. Recording it would currently be a no-op anyway, since performInsert()'s restore loop would assign the same expression back. There is a genuine latent gap here — a SpatialExpression assigned to a point attribute never becomes a Point after save — but fixing that means recording the geometry behind the expression, which is a behaviour change and out of scope.
• Support/Utils globe ISO check — all 255 features in the bundled globe.json carry both ISO_A3 and ISO_A2 (verified); the guard protects against future data.
• Api/v1/OrderController::capturePhoto's empty-photo check — defensive after a validate() that already requires a non-empty photos array.
• Api/v1/OrderController::updateActivity's not-found guard — blocked on an upstream defect: core-api's Model::findByIdOrFail() calls a non-existent getModelNotFoundException(), so it raises BadMethodCallException instead of ModelNotFoundException and a missing order returns 500 rather than 404. Remove the annotation once core-api is patched. Details in this comment on #277.

Coverage impact

Metric	Before	After
Lines	99.42%	99.52%
Methods	96.69%	97.19%
Classes	81.38%	82.92%
Uncovered statements	198	165

The --fail-under=100 gate is still red: 165 statements remain, of which two are the annotated upstream-blocked lines and the rest are residue not yet classified.

Verification

Run on the host runtime (asdf PHP 8.4 with Xdebug), no Docker:

• composer test:lint — clean, 0 of 1140 files needing fixes
• XDEBUG_MODE=coverage COMPOSER_PROCESS_TIMEOUT=0 composer coverage:baseline — all suites green
• Per-file slice coverage confirming each newly reachable line actually executes
• php scripts/coverage-summary.php coverage/clover.xml --fail-under=100

Behavioural canaries specific to this change:

• Spatial write paths: ZoneControllerBordersAndSeamsTest (4/4) and ServiceAreaControllerBorderAndSeamsTest (3/3) snapshotted before and after, including a new polygon update round-trip that rewrites a border and reads the vertices back.
• Binding-level cast assertion verified to discriminate: fails on the old Casts/Polygon shape, passes on the new one.
• coordsToCircle: asserted the ring still opens and closes at the same coordinate with no duplicated interior vertex.
• Group-by-group: every pre-existing test still passes, since a behavioural diff on a supposedly-dead line would mean it was not actually dead.

Three tests that asserted the old behaviour were updated (the Null Island lookup, and two that documented the Casts/Polygon return shape). Each is called out in its diff with a comment explaining what changed and why.

Notes for review

• Four lines became genuinely reachable as a result of these fixes and are now covered rather than ignored: the invalid-coordinates error, insertFromMixed(GoogleAddress), and two guards whose callee can in fact return null (Utils::getPointFromMixed() returns null for an unresolvable place_*/driver_* id — worth knowing, it is easy to assume otherwise from the signature).
• Not included, flagged for separate consideration: Casts/Point returns a raw Point from its isCoordinates arm while its GeometryInterface arm wraps — the same class of inconsistency as the Polygon fix, left alone to keep this diff's blast radius contained.