perf: enforce pagination regression budget

[cervantesh]

Summary

• promote the merged pagination baseline job to a required regression gate
• enforce deterministic p99, peak RSS, and peak heap limits against machine-readable measurement output
• record three fresh merged-workflow hosted attempts, observed maxima, threshold headroom, and rebaseline policy
• upload both measurement and enforcement-result JSON files for failed or successful runs

Thresholds

• p99 latency: 12 ms (106.2% above the 5.819 ms hosted maximum)
• peak RSS: 134,217,728 bytes / 128 MiB (47.8% above the 90,787,840-byte maximum)
• peak heap: 33,554,432 bytes / 32 MiB (48.8% above the 22,546,576-byte maximum)

Baseline evidence

Three independent attempts of merged-main CI run 29459726382 passed on ubuntu-latest with Node v22.23.1. All used 3 warmups, 25 samples, the deterministic 100-page/10,000-item fake, and all four guard fixtures without live GitHub API access.

Local validation

• npm run verify
• npm run benchmark
• npm audit --audit-level=high
• npm pack --dry-run

Closes #45

[cervantesh]

Hosted enforcement evidence is available from CI run 29460075294 as artifact pagination-benchmark-results (artifact id 8360810438).

The artifact contains both pagination-benchmark.json and pagination-budget-result.json:

• p50 / p95 / p99: 3.863 / 4.738 / 6.046 ms
• peak RSS: 88,637,440 bytes
• peak heap used: 22,544,216 bytes
• 25 samples after 3 warmups
• all guard fixtures passed; liveGitHubApi: false
• deterministic enforcement result: ok: true, no violations
• limits applied: 12 ms p99, 134,217,728-byte RSS, 33,554,432-byte heap

Both test and required Pagination regression budget checks passed. PR remains unmerged.