Stabilize retained-heap qualification

[Ismail-elkorchi]

Summary

• replace adaptive heap convergence with one fixed full-collection sequence
• size retained parser cohorts to amortize fixed process-level parser allocation
• report and contract-test the exact collection and cohort policies

The previous eight-result medium cohort made an approximately 806 KB fixed V8/parser allocation large enough to produce a bimodal baseline. The revised cohort measures steady-state retained bytes per public result without changing fixtures, runtime heap caps, sample count, revision interleaving, comparison thresholds, or historical enforcement.

Validation

• 12 focused independent-baseline repetitions for both parser sizes
• npm run lint
• npm run typecheck
• node --test test/tooling/performance-policy.test.js
• npm run qualification:release

All parser and serializer comparisons pass against the immediate baselines and tagged v0.1.1; manifests remain at 0.1.1.