perf: measure guarded pagination bounds

[cervantesh]

Summary

• export named hard maximums for guarded pagination
• add deterministic fake 100-page/10,000-item measurement with warmups, repeated samples, percentiles, peak RSS/heap, and guard fixtures
• upload a nonblocking hosted baseline artifact from the offline-validated CI workflow
• document the measurement/decision split without inventing a hard budget before hosted evidence exists

Validation

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

Refs #45

A follow-up budget PR is intentionally deferred until comparable hosted samples justify a stable threshold.

[cervantesh]

Hosted baseline evidence is available from CI run 29459133318 as artifact pagination-benchmark-results (artifact id 8360481976).

GitHub-hosted Node 22 sample (25 measured runs after 3 warmups):

• p50: 4.020 ms
• p95: 5.615 ms
• p99: 6.430 ms
• peak RSS: 89,092,096 bytes
• peak heap used: 22,544,224 bytes
• all four guard fixtures passed

This is one hosted baseline, not enough evidence to select a stable CI budget. A separate follow-up budget PR remains required after comparable hosted samples establish variance and a defensible threshold.