add mimalloc as optional dependency

[pca006132]

mimalloc is faster than glibc on linux, and it is probably also faster than the default allocator on Windows and Mac.
For extreme examples such as Samples.Sponge4, the difference can be up to 12%.

While on linux we can use LD_PRELOAD to override the default memory allocator, I wonder if we should add mimalloc into our code as an optional dependency. This can allow our pypi build to use mimalloc and benefit from the performance improvement.

[elalish]

I don't have a strong opinion. Does it help any other examples much? Does it slow any down?

[pca006132]

Yes it helps in general. So far I don't see any slow down, but I have only tested on Linux. Would be helpful if others can test on Mac and Windows. Will propose a patch later for testing.