release: decide what a breaking change of a silent type (docs!:) should publish

[dcotelo]

Problem

Confirmed while reviewing the release process: doc-only changes correctly
publish nothing. .github/scripts/release-version.sh ranks feat a minor,
fix/perf a patch, ! a major, and "everything else — docs, chore, test, ci,
refactor — publishes nothing at all", and #32 (docs-only) published nothing,
riding into 0.14.0's notes through its promoted Unreleased entry. That part
works.

One edge is decided but perhaps not deliberate: a breaking change of an
otherwise silent type. docs!: currently publishes a major, because the !
is ranked before the type is consulted.

Proposed behavior

Pick one and make the rank table say so explicitly. Two defensible answers:

• Keep it. A ! means a breaking change whatever the type, and
  documentation that breaks a documented contract — removing a command from the
  docs, changing a config key's meaning — is a real break for users.
• Narrow it. Only types that can publish at all get to publish a major, so
  docs!: would be a no-op and a genuine break would have to be feat!: or
  fix!:. Makes the rank table read as "silent types are always silent".

Whichever way it goes, tests/test_release_version.sh should pin it, so the
next person reading cp_rank does not have to infer the intent.

Alternatives considered

Leaving it undocumented and unpinned. No behaviour is wrong today, which is why
this is a question rather than a bug — but the next edit to cp_rank could flip
it either way without anything noticing.

Additional context

Surface: CI and release automation (.github/scripts/release-version.sh,
tests/test_release_version.sh).

Worth knowing: the ! is checked first deliberately in the current code — the
comment there says a breaking change of an otherwise silent type "still has"
somewhere to go. So this may already be the intended answer, in which case the
work is a test and a sentence rather than a behaviour change.