CI: per-category parity thresholds in test-parity/threshold.json

[TheHypnoo]

Replace the single 83% global threshold with per-domain thresholds (async, classes, modules, regex, JSON, Symbol, Proxy, fs, crypto, stream, …). CI fails when any category regresses, not just the global average.

Seed data: @proggeramlug has per-module diff counts across 5 sweeps (v0.5.713 → v0.5.910) ready to drop into threshold.json as the initial baseline. This sub-issue composes with #812 (per-module behavioral matrix) — that matrix is the natural data source for these thresholds; this issue is just the regression-gate layer on top.

Part of #793.

[andrewtdiz]

I’m taking this as a focused parity-threshold gate cut. Planned scope: add a script that reads test-parity/reports/latest.json plus test-parity/threshold.json, enforces the existing global minimum and new per-category/module minimums, emits a markdown/JSON summary for CI artifacts, and wire it into the existing parity workflow. Non-goals: changing runtime parity tests, changing known-failure semantics, or setting aggressive thresholds beyond current configured baselines.

[andrewtdiz]

Opened draft PR #4388 for this: https://github.com/PerryTS/perry/pull/4388\n\nScope: adds , expands with per-category defaults/exceptions, and wires the existing parity CI threshold step to emit/enforce category threshold reports while preserving the global 83% gate.

[andrewtdiz]

Correction to my previous issue comment: opened draft PR #4388 for this slice:

#4388

Scope: adds scripts/parity_threshold_gate.py, expands test-parity/threshold.json with per-category defaults/exceptions, and wires the existing parity CI threshold step to emit/enforce category threshold reports while preserving the global 83% gate.