Turbopack: Improve compaction in Persistent Caching (#80860)
### What?
Refactors the compaction algorithm with the following improvements.
* Multiple ranges can be compacted in parallel
* Instead of going for the largest merge it looks for an "optimal" size and runs multiple in parallel
* Compaction considers duplicated size and tries to reduce it
* Compaction works on recent SST files first as they tend to be less compacted yet and potentially contains more duplication.
* Limits the parallel compactions to CPUs / 4 during the build and CPUs / 2 after the build
