Ref T12319. With large datasets, the computation of which packages own paths in a revision is needlessly slow.

Improve performance through caching:

• Cache which paths belong to each repository.
• Cache the split fragments of each path.
• Cache the path fragment counts.
• Micro-optimize accessing $this->path.

• Used bin/lipsum to generate 4,000 packages with 150,000 paths.
• Created a revision affecting 100 paths in phabricator/ (these paths mostly overlap with bin/lipsum path rules, since Lipsum uses Phabricator-like rules to generate paths).
• Before optimizations, this revision spent about 5.5 seconds computing paths.
• After optimizations, it spends about 275ms.