fix(queue): guard ISSUE_CLOSURE solvedByPr write with state = CLOSED

[rsnetworkinginc]

Summary

The async ISSUE_CLOSURE fetch job can write issues.solved_by_pr onto an issue
that has already been reopened. handleIssueClosure reads issue state once, does a
slow GraphQL timeline fetch (fetchIssueClosingPr), and then writes solvedByPr
unconditionally. If an issues.reopened webhook lands during that fetch, its
handler upserts state = OPEN and clears solved_by_pr, but the in-flight job then
clobbers that clear and leaves an OPEN issue carrying a solver-PR link.

This corrupts issue-discovery scoring inputs: an OPEN issue can be counted as solved
(or skew credibility ratios) until a later webhook/backfill happens to correct it.

Root cause

packages/das/src/queue/fetch.processor.ts — handleIssueClosure() reads
issue.state once, then performs the slow GraphQL fetch, then writes:

await this.issueRepo.update({ repoFullName, issueNumber }, { solvedByPr });

with no re-check of state — a classic read-then-write race against the synchronous
reopen path in IssueHandler.

Fix

Guard the final write with state = 'CLOSED' in the update criteria, so the write
becomes a no-op if the issue was reopened while the GraphQL fetch was in flight:

await this.issueRepo.update(
  { repoFullName, issueNumber, state: "CLOSED" },
  { solvedByPr },
);

This is the UPDATE … WHERE state = 'CLOSED' direction suggested in the issue. It is
a single-statement conditional update (no extra round-trip, no schema change) and does
not touch the reopen webhook path. Distinct from #188/#189 (frozen closed_at
attribution) and #24 (the reopen clear itself) — this closes the concurrency window
that lets attribution reappear after that clear.

Tests

Added regression coverage in fetch.processor.spec.ts, driven through the public
process() entry point with an in-memory Issue repository whose update honours the
criteria like SQL WHERE:

• reopen race — the closing-PR fetch mutates the row to state = OPEN mid-flight;
  asserts the issue stays OPEN with solved_by_pr = null and that the final write is
  guarded by state: "CLOSED". This test fails on the pre-fix code (the OPEN issue
  gets solved_by_pr = 123) and passes with the guard.
• happy path — an issue that stays CLOSED still gets solved_by_pr written.

Verification

• npm run lint — clean
• npm run format:check — clean
• npm test — 27/27 pass (RED→GREEN on the new race test)
• npm run build — clean

Closes #199