Speedup analysis of arrays containing closures

[staabm]

alternative implementation to #4680

degrade analysis precision to a callable type

refs phpstan/phpstan#13933

---

before this PR

php bin/phpstan analyze ../phpstan-performance-issue-reproduction/src/index_bad.php --debug -vvv 

Elapsed time: 16.15 seconds
Used memory: 174 MB

after this PR

php bin/phpstan analyze ../phpstan-performance-issue-reproduction/src/index_bad.php --debug -vvv 

Elapsed time: 2.24 seconds
Used memory: 90 MB

---

before this PR

➜  phpstan-src git:(2.1.x) ✗ time vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php --filter testBug13933
PHPUnit 11.5.46 by Sebastian Bergmann and contributors.

Runtime:       PHP 8.3.29
Configuration: /Users/staabm/workspace/phpstan-src/phpunit.xml
Random Seed:   1767176262

.                                                                   1 / 1 (100%)

Time: 00:06.510, Memory: 92.00 MB


OK (1 test, 1 assertion)
vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php --filte  6.41s user 0.17s system 99% cpu 6.580 total

after this PR

➜  phpstan-src git:(2.1.x) ✗ time vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php --filter testBug13933
PHPUnit 11.5.46 by Sebastian Bergmann and contributors.

Runtime:       PHP 8.3.29
Configuration: /Users/staabm/workspace/phpstan-src/phpunit.xml
Random Seed:   1767176352

.                                                                   1 / 1 (100%)

Time: 00:00.906, Memory: 80.50 MB

OK (1 test, 1 assertion)
vendor/bin/phpunit tests/PHPStan/Analyser/AnalyserIntegrationTest.php --filte  0.81s user 0.16s system 99% cpu 0.979 total

[phpstan-bot]

This pull request has been marked as ready for review.

[staabm]

this PR will not close phpstan/phpstan#13933 because the reproducer was not representable (see discussion in the linked issue)

--

after rebasing this PR onto the yesterdays changes the test got 1 second slower, which is caused by d8decc8

[ondrejmirtes]

the test got 1 second slower, which is caused by

I reverted it because it's really tricky to remember there's a cache when changing something about OptimizedDirectorySourceLocator.

I'd accept revert of the revert (so that the cache is on disk even for tests), but OptimizedDirectorySourceLocatorTest should have a special .neon file where it's memory cache again.

[ondrejmirtes]

Thank you!