[clang] [analyzer] Improve cache locality by using separate allocators (PR #138295)

[via cfe-commits]

tigbr wrote:

@steakhal As we have discussed in person, unfortunately, I could not reproduce the big improvements for the three problematic translation units.

On the bright side, however, I was able to reproduce an average improvement between 0.5% and 1%. This is in the ballpark that I have measured for the vim project originally. I could also measure a decrease in the number of LLC misses with perf (with and without ASLR). Here are the results of these measurements: [perf-LLC-miss-measurements.zip](https://github.com/user-attachments/files/21024564/perf-LLC-miss-measurements.zip)

FYI: @NagyDonat

https://github.com/llvm/llvm-project/pull/138295