Decide whether "lib.rs, then Cargo.toml, then STATUS.md" is the repository-wide authority order for a documented claim

[justin13888]

A maintainer decision, raised by PR #526 (#425) and deliberately not taken there.

What #526 established, and how far

#526 corrected the README crates table. One row (gamut-dsp) had been derived faithfully from
crates/gamut-dsp/STATUS.md:67, which lists the jpeg module under "Deferred / tracked
follow-ups" — while crates/gamut-dsp/src/lib.rs:43 declares pub mod jpeg and
crates/gamut-dsp/src/jpeg/dct.rs ships fdct8x8/idct8x8. The row was true to the file and
false about the crate.

#526 settled it for its own deliverable: where a crate STATUS.md contradicts its lib.rs,
lib.rs wins — a claim must be true of the crate, not merely faithful to a file. That rule now
appears in three places, all of which #526 owns: the guard script header, the failure message a
contributor reads when the guard fires, and the mise.toml task description
(tooling/check-readme-crates/run.sh, mise.toml [tasks.check-readme-crates]).

The open question

Generalising that ordering to every documented claim in the repository — AGENTS.md, each
crate README.md, docs.rs module docs, Cargo.toml descriptions, STATUS.md — is a larger claim
than one README table established, and it is a maintainer call, not a lane one. So #526 put the
rule where the tooling speaks and filed this instead of editing AGENTS.md.

Reasons to take it repository-wide:

• The same defect keeps recurring outside the crates table: STATUS.md drift: gamut-dsp lists a shipped module as deferred, gamut-avif contradicts itself #545 (two STATUS.md files assert
  what their crates contradict) and gamut-av1 and gamut-avif publish crate descriptions that their own source contradicts #560 (a Cargo.toml description and a module doc assert what
  their own source contradicts) are both instances, found in one pass.
• Without a stated order, two documents disagreeing is a stalemate: each cites the other.

Reasons to hesitate:

• STATUS.md is forward-looking in a way lib.rs is not: it records intent, disposition and
  what is deliberately out of scope. "lib.rs wins" is right for "does this exist", and says
  nothing about "is it finished" or "is it supported" — a pub item can exist and still be
  unstable.
• The rule as written is about capability claims. Making it universal risks reading it as
  "documentation is descriptive only", which is not what STATUS.md is for.

Scope

• Decide whether the ordering is repository-wide, and if so with what carve-out for intent and
  disposition claims.
• If adopted, record it once in AGENTS.md and let the three sites docs(readme): correct the crates table against each crate STATUS.md, and gate it #526 added cite it rather than
  restate it.
• Consider what is mechanically checkable. STATUS.md drift: gamut-dsp lists a shipped module as deferred, gamut-avif contradicts itself #545 already notes one: a STATUS.md listing as
  deferred a module its lib.rs declares pub is a grep, not a judgement call.

Acceptance

Either AGENTS.md states the authority order with its limits, or a written decision records why
the rule stays local to the crates-table guard.

Refs #425, #526, #545, #560.