fix(engine): an ordScope bound shorter than the index key bounds by prefix

[russimicro]

Clipper / DBFCDX treat a scope bound shorter than the index key as a prefix: every key beginning with it is inside the scope. Three places assumed the bound was always full width, and together they emptied the scope instead.

1. Table::key_in_bottom_scope_ compared the full-width key against the bound

return key <= *order_->scope().bottom;

Every extension of a prefix sorts after the prefix. With a CON+DOC+STR(SEQ,6,0) tag (16 bytes) and a CON+DOC bound (10), every row fell past the bottom — the scope came out empty, where DBFCDX over the same table returns that document's lines. key_in_top_scope_ was right by accident (an extension is >= the prefix); it is made symmetric so a future edit cannot break one without the other.

2. AdsSetScope padded the bound out to key_length

That padding was added so an unpadded bound on a single-field tag would work — rddads sends OrdScope() values at hb_itemGetCLen(), i.e. trimmed. But it is exactly what breaks the partial-key case that shares the path: the bound becomes "CON+DOC" + 6 spaces, which every real key sorts after.

With prefix comparison the padding is unnecessary for both cases, so it is gone. An over-long bound is still trimmed to key_length.

3. CdxIndex::count_scoped_keys compared key_size_ bytes against the bound

Once bounds are no longer padded, that would also read past a short bound's buffer. It now compares min(bottom.size(), key_size_).

Test

abi_scope_partial_key_test — partial bound as top+bottom (selects the whole document), bottom-only, top-only, full-width bounds (exactly one line), a partial bound matching nothing (empty scope), and no scope at all.

Plus a second case covering what the padding in (2) was protecting: a trimmed bound on a single-field C(8) tag must still select its row — now by prefix comparison instead of by padding the bound.

Real-data validation

Harbour 3.2 + rddads, production table with a 16-byte tag. ADS now matches DBFCDX in all four combinations of RDD × SET DELETED, for the scoped walk, for a partial seek inside an active scope, and with a filter under SET OPTIMIZE ON:

[ADS   ] DELETED OFF/ON  ordScope parcial -> 1 row (expected 1)   OK
[DBFCDX] DELETED OFF/ON  ordScope parcial -> 1 row (expected 1)   OK

Before the change both ADS rows returned 0.

Suite: 1234/1240 — the 6 failures are the pre-existing Remote: locking cases in abi_pritpal_lock_test (embedded TCP server), identical with and without this change in our environment.

Found running a Harbour/FiveWin ERP on OpenADS, while checking the other consumers of the key-comparison primitive touched by #154. Independent of that PR: this one applies cleanly on its own.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[FiveTechSoft]

Integrated into main as part of the PR stack merge (9e7ac41). Cherry-picked onto post-CI-fix main, conflict-resolved (CHANGELOG / CMakeLists), and verified locally. Thank you @russimicro.