test: add performance regression guards

[zefutoff]

Summary

• add real-engine performance regression coverage
• verify repeated analyses reuse the initialized QuickJS runtime without reloading Grammalecte assets
• add tolerant timing budgets for cold start, sentence analysis and paragraph analysis
• keep the existing opt-in characterization benchmark
• add a dedicated make performance-check target
• run performance regression checks in the existing CI quality job
• document the measured development baseline and regression budgets
• complete the Phase 4 performance-regression roadmap item

Regression budgets

• cold first correction: ≤ 3000 ms
• sentence median: ≤ 75 ms
• sentence p95: ≤ 200 ms
• paragraph median: ≤ 150 ms
• paragraph p95: ≤ 400 ms

The thresholds are intentionally wider than the current development baseline to detect substantial regressions without making CI sensitive to normal host-load variation.

Development baseline

Three representative local runs measured:

• cold first correction: 809.54–828.45 ms
• sentence median: 13.48–13.84 ms
• sentence p95: 14.94–16.45 ms
• paragraph median: 31.89–32.23 ms
• paragraph p95: 32.85–34.35 ms

A final regression run measured:

• cold first correction: 809.58 ms
• sentence median: 13.90 ms
• sentence p95: 15.52 ms
• paragraph median: 32.14 ms
• paragraph p95: 34.01 ms

Validation

• make check
• make performance-check
• make assemble
• git diff --check
• final APK does not request android.permission.INTERNET

Roadmap

Completes:

• Add performance regression tests