ci: skip redundant exact-hit saves for native app caches

[zackees]

Summary

Refs #1286; extends merged #1381's Ubuntu exact-hit guard to measured native app cache lanes.

Windows, macOS and Flatpak compiler-cache consumers restored exact keys, then compressed the same stores before immutable-key reservation failures. Add restore IDs and exact-hit save predicates to those three app jobs. Existing event/branch/ccache-success conditions remain intact; misses and compatible fallback restores can publish new keys. Restore/build/test paths and cache writer policy do not change. No new workflow or Tier 2 recovery changes (#1362).

Actual evidence

Same-revision full consumer run 39090099191 passed. Existing-key save step durations:

• Windows arm64 job 117564891081: 8 s; x64 job 117564891135: 4 s. Actual compiler hits 271/271 and 297/297; statistics were zeroed before consumers.
• macOS arm64 job 117564891140: 7 s; x86_64 job 117564891103: 6 s. Exact archive restores are proven; compiler hit counts are absent, so no macOS cache effectiveness claim.
• Flatpak job 117564891083: under one API-rounded second; redundant save warning observed. Its printed compiler counters are cumulative (ci: distinguish cumulative Flatpak counters from job performance #1383), not job deltas.

These are individual avoided-work opportunities, not measured merge-queue speedups. The source revision differs after this change; no additional full workflow was dispatched just to verify the guards.

RED → GREEN

• RED b8c414ca4: all three job cases fail due to missing restore IDs/exact-hit save guards.
• GREEN: uv run -m unittest ci.test_native_cache_exact_hit ci.test_ubuntu_cache_preserve ci.test_ci_toml ci.test_ci_tests_stdlib_only ci.test_release_native_cache passes all 16 tests. Covers matching restore/save keys, all existing guard conditions, Ubuntu fallback compatibility, declared families and standalone test compatibility.
• git diff --check passes. Local RED and separate GREEN pushed together under coordination policy.