docs: name both new IRR-CA anchors in the SoundDispatchOutcome docstring

[jleinenbach]

What

Adds one paragraph to the class docstring of SoundDispatchOutcome in
const.py, pointing at the two anchors this type is the carrier of:

• IRR-CA-SOUND-FAILURE-CLASS for the per-member "may a caller arm a push
  cooldown" table;
• IRR-CA-STOP-BREAKS-SELF-INFLICTED-COOLDOWN for the stop-path qualifier.

Documentation only. No behaviour change, no signature change, no test change.

Why

SoundDispatchOutcome exists because a bool cannot carry the state space, and
the rule that only TRANSPORT_FAILED may arm a push cooldown is the whole
point of the type. That rule and both of its anchors lived in
docs/PLAY_SOUND_ARCHITECTURE.md and in comments in coordinator/locate.py
and api.py, but not in the docstring a caller actually reads when they hover
the type. This repository already treats that shape as a defect class: see
tests/test_stop_sound_correlation.py::test_api_docstrings_declare_no_ring_confirmation,
which pins the same discipline for api.py ("a boundary that lives only in a
design document gets re-discovered as a bug").

This is the last open item of the DoD of AP-7 in the plan
PLAN_GFMY_SOUND_FAILURE_CLASSIFICATION. AP-1 to AP-7 shipped with #1262; only
this docstring reference was missed.

Measured

• Anchor gate (every IRR-CA- anchor cited under custom_components/ resolves
  in the architecture document): no MISSING line, before and after.
• git diff --name-only origin/1.7 → custom_components/googlefindmy/const.py
  only, 11 added lines, nothing removed. README.md and strings.json
  untouched, as the DoD requires.
• ruff check and ruff format --check: clean.
• pytest tests/test_stop_sound_correlation.py tests/test_sound_dispatch_contract.py tests/test_coordinator_sound_uuid.py tests/test_coordinator_locate_basics.py
  → 123 passed.

Review note

The first draft of this paragraph was wrong in a way worth recording: it tied
"a play cannot silence the stop its cancel key exists for" to
_note_stop_transport_problem_without_extending(). That function does
something else. The break-through happens at the readiness gate
(coordinator/locate.py, the used_own_fresh_key and time.monotonic() < self._push_cooldown_until branch); the function only keeps a running window
from being restarted by a stop that already broke through, which is the
separate "it cannot feed itself" property. The paragraph now states both, in
the order the code has them, and no longer claims the window was necessarily
armed by the very play that stored the key -- the code says that is the common
case, not a proven one.

[jleinenbach]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Another round soon, please!

Reviewed commit: f3b70f88cc

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!