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

[Balazs Benics via cfe-commits]

steakhal wrote:

I had a look at the spreadsheet and the numbers for `linux-topro.c` looks really impressive.
However, in general, we need further investigation about why are we not getting faster e.g. in the `clang-MSP430ISelDAGToDAG.cpp` case. The direction seems right though.

(FYI it usually better to take the minimums of the different analysis attempts than the avg).

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