docs: the presentation standard states what is true, not what was planned

[JulienMartel]

What

docs/cli-presentation.md's Order of work table and the run of "What N
shipped" sections behind it are gone. Every row was done once hausfold/haus#568
landed the last three painters, and this repo's own rules say a finished plan
stops being a doc and that a doc states what is true now, not how it got that
way.

The same content returns in the present tense, minus the history and the PR
numbers, as three sections:

• Who draws through it — the caller inventory (scruff imports the package;
  bench, haus.sh/haus-show.sh drive the coprocess; the statusline and
  image preview take roles only; lane-open.sh injects HAUS_UI_SH into the
  snippet another shell runs), the two escape bans of different strengths, the
  one exception named inside the stricter one, and the three role collisions.
• Rules a caller has to meet — the record protocol, one coprocess per
  command opened lazily and staying dead once dead, the descriptor a background
  painter must duplicate, the sudo repaint rule, which-stream-is-a-property-of-
  the-command, one colour precedence, forcing the tty gate where colour is right
  without one, the UI_SH sentinel collision, bash 4+, the width probe that can
  kill its caller, and where the signal policy lives.
• Why the fallback lives in snug — generated from one list, shipped beside
  the binary, reachable by a haus user with no workshop checkout.

Where the standard stops is untouched — decisions, not a backlog.

bench (two comments), test/bench.bats (one) and AGENTS.md's routing row cited
the table by step number; they now cite the section that holds the fact. The
routing row drops its "one row is still open (10)" warning.

Why

The table was the durable record of what was left. There is nothing left, so it
was a record of history in a file whose whole job is to be the current contract —
and four files cited it by a number that was about to stop resolving.

Verify

bats test/bench.bats — 186/186 (comment-only changes to bench; the suite is
run because it greps the script's own text in places).

Sweep confirming the claim: across the family, the only shell files still
holding an SGR escape are bootstrap.sh, haus-activate.sh,
build-golden-vm.sh, trill/scripts/dev-install.sh and this repo's
script/issue-labels.sh + script/probes/*.sh — every one of them named in
Where the standard stops — plus snug/share/ui.sh, which is the painter,
and the statusline's TINT_FABLE, which the doc names.

Watch out

• A clean-context assurance pass caught three wrong claims in the first cut, all
  fixed before this PR: the two bans are not one ban (the painters' bans SGR
  colour only, which is why image-preview.sh may ship a DECTCEM cursor
  escape), the statusline has three collisions not two (ADD/PR_OPEN→ok
  and DEL/PR_CLOSED→err is the third), and "no hardcoded index survives"
  holds for user-driven surfaces only.
• test(phase-painter): cite the standard's sections, not its step numbers haus#569 does the same de-citation in test/phase-painter.bats.
  Independent — neither PR depends on the other's order.
• bench overlap: clean against every lane and origin/main; one lane is
  elsewhere in bench and test/bench.bats, no shared region.