Make ResultCacheManagerCollectedDataTest independent of the second boundary - #6898

ResultCacheManagerCollectedDataTest fails intermittently on 2.3.x. On 92a35f5 it failed in both Tests (PHPUnit 11.x (8.2, ubuntu-latest)) and Compile PHAR (Run with Turbo Extension (ubuntu-latest, 8.2, nts, make tests)), in testPartialSaveEqualsFreshSaveOfMergedData. The only difference was the fileStat entries in dependencyGraph: the partial save had a/c signed and the fresh save had none.

Baseline and first bad commit

• Last all-green baseline: 524520f. Every workflow passed except Benchmark, which also fails on 9c8ed05.
• 9c8ed05 added the test. Its Tests run passed by luck of timing. 92a35f5 only touches an unrelated fixture and fails on the same timing. Locally, at an unchanged 92a35f5, the test failed in 1 of 30 runs (once also in testWarmSaveCopiesEveryEntryByteForByte).

Root cause (test-only)

• saveFullAnalysis() restored with one ResultCacheManager and processed/saved with another, so the "fresh" save never had the signatures restore() recorded and always wrote N. In production, the same manager restores and saves.
• FileStatSignatureReader leaves out signatures of files modified in the second the read began. So the partial and warm saves record signatures only when a second boundary has passed since the files were written.

The fix restores and saves with the same manager and waits until the files are in the past before saving, the same way DirectoryWalkerTest and OptimizedDirectorySourceLocatorFactoryTest do. No src/ change.

Verification

• A temporary probe that forces a second boundary between the saves fails 3/3 on the old test, fails with either part of the fix alone, and passes 5/5 with both parts.
• 40/40 unforced runs passed. tests/PHPStan/Analyser/ResultCache and tests/PHPStan/File pass. make cs passes, and PHPStan reports no errors on the file.
• The change is test-only, so the downgraded PHAR is unaffected.