A refused dispatch reads as a missing one, so the tool advises undoing it

[gHashTag]

tri gates unmeasured prints dispatch: NO beside the advice "add
workflow_dispatch: first" for every workflow that has no dispatch. It cannot
tell a workflow that is MISSING one from a workflow that REFUSED one.

#3325 removed the dispatch from release.yml because it could only ever fail:
every job keys off github.event.release.tag_name, which is empty on a
dispatch, so preflight takes its *) branch, prints "Tag '' names no product",
exits 1, and every publishing job is skipped. The reason was written in a YAML
comment. This tool does not read YAML comments.

Measured on #3325's head, before this change:

2026-08-28  NO   -   Release Pipeline  (gHashTag/t27)
...
`dispatch: NO` means the reading cannot be taken on purpose
-- add `workflow_dispatch:` first.

So the tool tells the next reader to put a dispatch back in front of cargo publish and npm publish against live registries, where a version is
permanent — undoing a decision made one pass earlier, for a reason the tool
never sees. The only thing standing between that advice and a re-added
dispatch is a comment addressed to humans.

A refusal has to be recorded where the tool looks. This is the shape of
"a new cron cycle must not break the previous one's work": the guard against
undoing a decision cannot live only in prose.

[gHashTag]

Closed by #3356, merged as c8b48a3.

Verified on origin/master, with a tri built from master itself — not from the merge message:

2026-08-28  refused    -         Release Pipeline  (gHashTag/t27)
...
`dispatch: refused` means the workflow declined one ON PURPOSE and
says so with a `# tri:no-dispatch` comment. Do not add one.

• has_dispatch returns three states (Yes / No / Refused); .github/workflows/release.yml carries the # tri:no-dispatch marker (1 occurrence on master).
• Both dispatch tables now explain refused in their footnote — the second one was caught by a structural test, because my first patch of it was a silent no-op.
• cargo test -p tri --bin tri dispatch_refusal: 6 passed, 0 failed.

A present workflow_dispatch: still wins over the marker, so a stale comment cannot hide a real dispatch — asserted by a_present_dispatch_wins_over_a_stale_marker.