feat: switch issue-comments fetch to GraphQL with isMinimized filter

[twistedmelonman]

Summary

Coordinating change to smartwatermelon/github-workflows#83 (tracking issue: #82). That upstream PR taught the CI claude-blocking-review workflow to call GraphQL minimizeComment(classifier: RESOLVED) on its own PASS comment after posting it, so PASS comments end up collapsed in the PR thread (audit trail preserved, visual noise reduced).

However, scripts/post-push-status.sh was still fetching issues/comments via the REST API, and the REST endpoint does not return isMinimized — that field is GraphQL-only. The result: the scraper kept surfacing those minimized PASS comments as FINDING lines, polluting the post-push loop with phantom findings.

What changed

• New _fetch_issue_comments_gql helper that:
  • Issues a GraphQL pullRequest.comments(last: 100) query selecting body, createdAt, isMinimized, and author { __typename, login }.
  • Uses --jq to filter out minimized comments (select(.isMinimized == false)) and reshape GraphQL nodes into the REST-compatible JSON shape (user.login, created_at, body, path, line) the existing python parser already expects.
  • Re-appends [bot] to Bot-typename logins (GraphQL strips it), so BOT_PATTERN (claude\[bot\]|sentry\[bot\]|coderabbit\[bot\]) keeps matching.
  • Falls back to [] on any failure (matches _safe_api_array semantics).
• Replaces the single _safe_api_array "repos/.../issues/<N>/comments" call with the new helper.
• No python parser changes — the jq projection makes the GraphQL output drop in transparently.

Why pulls/comments stays on REST

Pulls/comments are inline diff review comments — they're typically real findings (e.g., from sentry[bot]), they're not subject to the new minimization workflow, and the REST shape works fine. Switching that path too would be churn without benefit.

Test plan

• bash -n scripts/post-push-status.sh — clean
• shellcheck -S info scripts/post-push-status.sh scripts/tests/test-post-push-status.sh — zero findings
• bash scripts/tests/test-post-push-status.sh — 17/17 pass (added 1 new assertion: assert_not_contains "minimized PASS comment dropped" "VERDICT: PASS"). The test mock now extracts --jq from the call's argv and pipes the raw GraphQL response through the real production filter via jq -c, so the test exercises the actual projection (Bot→<login>[bot] reshape + minimized drop) rather than baking that behavior into the fixtures.
• Smoke test against PR refactor(run-review): parallelize reviewers, CLI preflight, cache versioning, remove dead detection #128 (real-world: contains a minimized claude[bot] PASS comment): baseline produced 3 findings (2 sentry[bot] inline + 1 claude[bot] PASS); after fix, produces 2 findings (PASS correctly skipped, real findings preserved).
• Smoke test against PR feat: add commit message context to chunked code reviewer #149 (un-minimized claude[bot] PASS): still surfaces correctly — proves Bot login reconstruction works and we don't break the un-minimized path.
• Pre-commit + pre-push reviewers (code-reviewer, adversarial-reviewer, codebase reviewer): all PASS.

🤖 Generated with Claude Code

[claude]

No blocking issues found. The GraphQL helper correctly reshapes PR comment nodes into the REST-compatible shape the downstream Python parser expects, the [bot]-suffix reconstruction matches BOT_PATTERN, and all failure paths (network error, malformed response, non-array result) fall back to [] — same behavior as the old REST call. The comments(last: 100) limit is at least as broad as the default REST page size, so no regression there. Test mock updates accurately reflect the new GraphQL code path.

VERDICT: PASS