[justin13888]

Release PR #820 fails Format & Metadata: check-readme-crates refutes the crates-table tokens for gamut-isobmff (v3 → 4.0.0), gamut-avif (v2 → 3.0.0) and gamut-exif (v2 → 3.0.0). The guard held each row's vN token and the pinned gamut = "X.Y" Usage snippets to the manifests, but release-plz bumps only the manifests. So every major bump, and every pre-1.0 minor bump of the umbrella, turns the release PR red.

This PR originally fixed that by rewriting the claims on the release branch with a script that parsed the guard's output. That approach kept the duplication and added a second parser. This version removes the duplication instead: the README states no version, so a release has nothing to update.

Change

• Crates table (README.md): Status cells state disposition (stable, placeholder, an issue) and no version. Planned releases are written 1.0 ("1.0 under Release gamut-core v1 #177"). The section intro and the Releases note no longer refer to versions in the table.
• Usage snippets (README.md, crates/gamut/README.md): cargo add gamut -F … replaces the pinned [dependencies] block. Cargo resolves the newest release and writes the caret requirement, so the snippet cannot go stale. That still covers README: the Usage snippet pins gamut = "0.1", three months behind the current release (supersedes #523) #559.
• check-readme-crates/run.sh:
  • It refuses a vN/vN.M token anywhere in ## Crates, in rows and prose alike, instead of comparing it to cargo metadata.
  • In either README, it refuses a gamut = … line in any code block and requires a cargo add gamut line in one.
  • The manifest-comparison code is deleted.
• fixtures.sh: the version and snippet fixtures are replaced to match these rules: tokens in a row or in prose, pins in inline-table, bare-string and nested-fence form, a deleted or unfenced snippet, cargo add gamut-png not counting, and the umbrella README's own copy.
• release.yml is unchanged from master.

Verification

• tooling/check-readme-crates/run.sh passes on the tree.
• fixtures.sh passes 81/81 under both awk and gawk --posix.
• Simulated chore: release #820: with gamut-isobmff at 4.0.0 and gamut at 0.5.0, the guard still passes with no README edit.
• shellcheck -S warning is clean on both scripts, and convco check is clean.

Once this merges, the release-pr job regenerates #820 and the check passes with no extra commit on the release branch.

Closes #954

Decisions taken

• 596a50a (pipefail SIGPIPE race in run.sh). grep -q exits at its first match. Under set -o pipefail the writer upstream of it then takes SIGPIPE, the pipeline fails, and the guard intermittently said a real crate was not a workspace crate. That was measured at 1 in 2000 calls on a loaded host, and the fixture battery hit it on "a cell naming a real crate". Both grep -q pipelines in run.sh now read a here-string instead of a pipe. The fix belongs in this PR because it made this PR's own fixture battery flaky.
• 55c29f7 (the same race in fixtures.sh). The battery's fragment assertion (printf '%s\n' "$out" | grep -qF) ran under set -uo pipefail and had the same race. It could fail a fixture whose output did contain the fragment. It now reads <<<"$out", the same form as run.sh. I searched both scripts for any other | grep -q and found none.
• 9726cd6 (umbrella docs.rs usage example). crates/gamut/src/lib.rs still pinned gamut = { version = "0.1", ... }, but the umbrella is at 0.4. It now shows cargo add gamut -F avif,jxl, the same form as both READMEs. I searched every crates/*/src module doc for a pinned gamut* = dependency line and found only this one. This PR does not extend check-readme-crates to read lib.rs. The guard's contract is the two READMEs, and the example now states no version, so it has nothing that can go stale.
• Pre-existing check-spec-clauses failure (ci: spec-clauses mise tasks use set -o pipefail under sh, which is dash on ubuntu-latest #962). Format & Metadata also failed on master: the spec-clauses mise tasks ran under a non-bash shell. ci: run the spec-clauses mise tasks under bash #963 fixed that and has merged. 7127370 merges origin/master into this branch so the check runs against the fix. The merge commit has a Conventional Commit subject because the commit-msg hook rejects git's default merge message.
• Paths beyond the Change list. mise.toml: only the check-readme-crates task description changes, so it matches the guard's new rule (Usage adds the umbrella with cargo add and pins no version). crates/gamut/src/lib.rs: see 9726cd6 above.
• Local gates at 9726cd6. These passed: tooling/check-readme-crates/run.sh, fixtures.sh (81 passed, 0 failed), shellcheck -S warning on both scripts, cargo clippy -p gamut -- -D warnings, RUSTDOCFLAGS="-D warnings" cargo doc -p gamut --no-deps, mise run fmt-check, and the pre-commit hooks (fmt, clippy, convco). In this checkout, fmt-check and the fmt hook only ran with __CARGO_TEST_ROOT set to the checkout. The checkout is nested under another clone of the repository, and cargo walked up from the excluded tooling/lcms2-oracle into that clone's workspace. CI does not have this nesting. CI was disabled for this run, so CI is still the final gate.