test: stabilize local UAT perf validation

[RyshMan]

Problem
Local post-release UAT for v1.3.4 was failing in quality_perf_bench_check because the UAT path was reusing the strict baseline-regression perf gate after a long local validation run, which made the outcome sensitive to local host variance.

Changes

• add a dedicated bench-uat-check target for local UAT that keeps absolute perf/context budget validation without reusing the stricter pre-release regression gate
• run the UAT perf block earlier and allow one retry for that UAT-only perf step
• document the separation between pre-release perf regression validation and local UAT perf validation

Validation

• ./gait doctor --json
• make prepush-full
• make bench-uat-check
• GAIT_UAT_RELEASE_VERSION=v1.3.4 bash scripts/test_uat_local.sh