Account for checkpoint reopen and report enrichment ownership

[chaoz23]

Summary

• Add versioned accounting for bounded checkpoint reopen/resume work and typed read failures, reusing the existing manifest, payload, decode, parse, and stable-ID validation passes.
• Add opt-in report-enrichment and created/reused report graph-materialization receipts.
• Surface lifecycle receipts in successful ordinary JSON and explicit report saves while preserving stable report/checkpoint IDs and persisted payloads.
• Keep search-session accounting and memory-policy paths on their existing builders.

Compatibility and safety

• Legacy checkpoint and report schemas remain readable; stable-ID algorithms and stored payload formats are unchanged.
• Receipts are additive, bounded, numeric/privacy-safe lifecycle evidence outside saved report and checkpoint payloads.
• JSON-stream still avoids the monolithic report graph; default human/Markdown and compile-failure behavior stays unchanged.
• Partial typed checkpoint-read failures retain only the bounded work reached before failure, without another payload read, decode, parse, hash, or graph traversal.

Verification

• Full serial suite: 351 total, 348 passed, 0 failed, 3 expected local sandbox socket skips.
• TypeScript build, syntax checks, and git diff --check passed.
• New lifecycle suites: 10/10 passed.
• Focused CLI lifecycle integration: 3/3 passed.
• Existing report/checkpoint regressions: 12/12 passed.
• Search-session and campaign-control regressions: 39/39 passed.
• Two independent final reviews found no remaining correctness, API, privacy, bounds, identity, payload, or policy-path blocker.
• Exact-source 50,000-state Intercept calibration (one excluded warmup plus four counterbalanced pairs): wall +1.81% and RSS -2.47% by ratio of medians; paired medians +2.14% wall / -2.75% RSS. All ten arms produced the same evidence hash, 297 endings, 29 knots, zero errors, and maxStates-only termination. The +11.21% wall pair is retained as an outlier. This passes the narrow <=5% median cell but is one-machine engineering evidence, not a portable performance or promotion claim.

Scope

This is a partial accounting slice under #216 and contributes component-byte attribution evidence toward #156, within parent #215. All three issues remain open.

It does not add checkpoint-v2 encoding, activate compaction, retention, allocation, eviction, or stopping policy, change search-session resource policy, publish a release, or restart an InkBench matrix.

Related: #215, #216, #156