Investigate performance of `wasmer` multithread

[breathx]

Problem to Solve

Once parallelism will be applied, it usually will use structure, where there is single wasmer instance, which works with multiple threads. I think that it has some overhead for safety interaction with the threads.

Possible Solution

Apply separate wasmer instance for each created thread. It will cost a time for compilation, but should increase performance. Also need investigate possibility of collisions on stored binaries caches on disk for this wasmers.

Notes

Discussed with @grishasobol. So @grishasobol asked to check this idea.