fix: Revalidate stale Issue relationship blockers

[idy]

Closes #44.

Problem

For a resumable session, the Issue stage sent the model only snapshotDiff(previous, current) plus previous_result. A wrong blocker from the first full review about metadata that never changes could not clear. On GizClaw/gizclaw#1361, Issues #1359 and #1360 already had native parent #1358, but the review said "the supplied full Issue metadata has parent: null". Because parent_number never showed up in the delta, every incremental review raised that blocker again.

Change

• pr-review/stages.mjs: added issueRelationships(), which returns the current issue_type, state, parent_number, sub_issue_numbers, sub_issues, blocked_by and blocking. Also added ISSUE_STAGE_INPUT_VERSION = 2.
• pr-review/run.mjs: every Issue-stage input now includes current_relationships and relationship_checks. The prompt tells the reviewer to re-check each previous blocker against these values and drop any blocker they contradict. It also says not to report body-vs-native parent consistency itself. The input version is part of the Issue stage identity, so cached Issue evidence gets re-checked once in an incremental turn. PR and code evidence are unchanged.
• issue-review/common.mjs: added issueRelationshipChecks(), which compares a body - Parent: #N line (or a same-repository owner/repo#N) with the native parent. Lines that name another repository or sit inside fenced code are ignored. analyzeIssue now adds a deterministic parent-relationship-missing or parent-relationship-mismatch blocker.
• README: documented the new behavior.

Validation

node .github/scripts/issue-review/test.mjs
node .github/scripts/pr-review/test.mjs
node .github/scripts/pr-readiness/test.mjs
node .github/scripts/review-request/test.mjs

All pass. New coverage:

• A previous result with a missing-parent blocker, a current snapshot with parent_number set, and only the body changed. The incremental input has a body-only delta, plus current_relationships.parent_number, a passing parent check, the previous blocker and the revalidation prompt. The blocker clears.
• The deterministic parent check: match, missing, mismatch, same-repository qualified reference, another repository, and fenced code.

The latest tag was not moved.

🤖 Generated with Claude Code

[github-actions]

❌ OpenAI PR Review: FAIL

Conclusion: Not ready. 1 readiness blocker must be resolved before merge.

Review checks

Check	Result
PR format	❌ FAIL
Issue design	✅ PASS
Code & plan conformance	✅ PASS

Scope: 8d532cc4ef · 3e6e534dfa..8d532cc4ef · full · 1 diff chunk

Usage: 1m 11s · 263,432 tokens · 80.6% cache hit · 5.989 credits

Blockers

• pr-linkage Pull request must natively close at least one same-repository Issue.

Summary

The complete diff's single chunk has no actionable findings or plan-conformance blockers.

Review metadata

• Commit: 8d532cc4ef
• Range: 3e6e534dfa..8d532cc4ef
• Mode: full
• Diff chunks: 1
• Model: gpt-5.6-terra
• Reasoning effort: medium
• Session: repo-1309321116-pr-43-v2
• Generation: c7fee32a3bbfabc0050656fbc53cd7238e10d3f4db35a48e102b66798348ba64
• Evidence: 8180415dd464e1b0972811908938337987737dfc4d4f9608e7d890fd22e00e1d

Totals

• Input: 259,326
• Cached input: 209,045
• Cache write: 50,248
• Output: 4,106
• Reasoning: 1,684
• Total: 263,432
• Estimated credits: 5.989
• Credit rate per 1M tokens: 62.5 input / 6.25 cached / 375 output

Token and cache details

Stage	Mode	Target	Time	Input	Cached	Hit	Output	Total	Credits
pr	deterministic	format rules	0s	0	0	0.0%	0	0	0.000
pr	full	pr	14s	34,826	20,340	58.4%	740	35,566	1.310
code	full	chunk 1/1	49s	153,937	120,146	78.0%	2,868	156,805	3.938
code	full	aggregate	8s	70,563	68,559	97.2%	498	71,061	0.740

[github-actions]

✅ OpenAI PR Review: PASS

Conclusion: Ready from the OpenAI review perspective. PR format, linked Issue design, and code/plan conformance passed with no actionable findings.

Review checks

Check	Result
PR format	✅ PASS
Issue design	✅ PASS
Code & plan conformance	✅ PASS

Scope: 8d532cc4ef · 8d532cc4ef..8d532cc4ef · incremental · 1 diff chunk

Usage: 26s · 254,537 tokens · 83.1% cache hit · 4.576 credits

Summary

No code changes occurred in this generation; the prior complete code review remains applicable, and the linked Issue introduces no plan-conformance blocker.

Review metadata

• Commit: 8d532cc4ef
• Range: 8d532cc4ef..8d532cc4ef
• Mode: incremental
• Diff chunks: 1
• Model: gpt-5.6-terra
• Reasoning effort: medium
• Session: repo-1309321116-pr-43-v2
• Generation: 9fad17845c35b46593df712f88ee1c3c3298915f939092c5cff604293a4f7af8
• Evidence: fda60ee57300e535984805950dad1d51317d763a4cb450d6f0f9e554a301eef5

Totals

• Input: 252,979
• Cached input: 210,132
• Cache write: 42,829
• Output: 1,558
• Reasoning: 314
• Total: 254,537
• Estimated credits: 4.576
• Credit rate per 1M tokens: 62.5 input / 6.25 cached / 375 output

Token and cache details

Stage	Mode	Target	Time	Input	Cached	Hit	Output	Total	Credits
pr	deterministic	format rules	0s	0	0	0.0%	0	0	0.000
pr	incremental	pr	9s	74,798	50,718	67.8%	474	75,272	2.000
issue	full	Issue #44	8s	84,102	78,011	92.8%	533	84,635	1.068
code	reused	chunk 1	0s	0	0	0.0%	0	0	0.000
code	incremental	aggregate	9s	94,079	81,403	86.5%	551	94,630	1.508