Use mimalloc allocator for released binaries

[ChayimFriedman2]

Speed them up (hopefully).

We can use jemalloc but it's harder to build on Windows. I didn't benchmark so maybe jemalloc is faster.

[matklad]

If mimalloc builds everywhere without problems, consider removing allocator option altogether and using mimalloc, always and unconditionally.

[Veykril]

Would be good to check whether it actuall is faster for our workload, also whether it has more memory overhead than the default or not

[ChayimFriedman2]

So: I benchmarked this. On Windows this saves 8 seconds in analysis-stats . (but the benchmark was quite noisy, so the difference might be bigger - there definitely is a difference, though), and saves 40mb (but the memory measurement is very imprecise without jemalloc). On Linux this saves 15 seconds, and I couldn't measure memory usage. I don't have a Mac. So, clearly worth it.

[lnicola]

I can't check right now, but IIRC mimalloc used to use a lot more memory :-).

[ChayimFriedman2]

I can't check right now, but IIRC mimalloc used to use a lot more memory :-).

I cancelled the merge. When will you be able to check it? On Windows memory usage even (slightly) improve.

[lnicola]

On Linux I get:

104.01user 2.96system 1:47.43elapsed 99%CPU (0avgtext+0avgdata 4643236maxresident)k # glibc
86.52user 2.85system 1:29.44elapsed 99%CPU (0avgtext+0avgdata 5156344maxresident)k # mimalloc
91.89user 3.07system 1:35.34elapsed 99%CPU (0avgtext+0avgdata 4654904maxresident)k # jemalloc