tooling(pm): H48 — a governed PR's ACCEPT verdict without its label or its 维护者速读 brief (#15956)

The seat's verdict on a governed PR is a HANDOFF, not a merge, and the handoff
is one act with two writes: `needs-user-decision` puts the PR in the
maintainer's review inbox, and a `## 维护者速读` comment makes reading it a
two-minute job. One act with two writes has exactly two ways to come apart, and
this report-only row is both of them — accepted and unfindable, or findable and
unread.

Population is the open PRs the sweep already lists, narrowed by the imported
`governedPathsIn` matcher over the changed-file pages H36 already fetched —
never a hand list of paths. Drafts are IN (a governed draft is the correct
terminal state of that regime and still half a handoff); merged and closed are
out. A governed PR with no verdict is CLEAN: it is not the maintainer's yet.

The brief and the verdict are read from the PR's own issue-comment thread, and
that is a NEW fetch class: `commentCache` holds card threads only (every
listing feeding it filters `!i.pull_request`). It is declared as H36 and H43
declare theirs, bought one thread per GOVERNED OPEN PR — 11 of 31 on the board
this landed against — and cached so a future PR-comment reader pays nothing. A
thread that failed or reached the page ceiling is UNJUDGED rather than short: a
partial read can only invent a missing half.

The `## 维护者速读(草稿)` section in a PR BODY is the dev's draft, written
before the review happened, and never satisfies the brief half — the row reads
comments only and says so in its own sentence.

Markers mirror `CLAIM_COMMENT_MARKER`/`RELEASE_COMMENT_MARKER` in three
properties (line-anchored `m`, optional leading blockquote, no `g`) and
deliberately not in the fourth: no `i` flag, for the reason
`RELEASE_COMMENT_MARKER` states — a rule that landed with no dialect behind it
starts at the canonical spelling and nothing wider.

Self-test 2341 -> 2414 cases, none dropped.


Claude-Session: https://claude.ai/code/session_019RfFHiRCSs3JXLK4cwcfox

Co-authored-by: Claude <noreply@anthropic.com>

Author: claude[bot]