Cache compiled native code produced from wasm modules

[kazimuth]

The main reason I'm interested in this is that it could speed up benchmarks run under callgrind by a lot. Note that it's possible to compile wasm modules while running callgrind, just quite slow.

This should probably wait until after we validate + merge wasmtime (since callgrind doesn't play nice with wasmer, but likes wasmtime.)

[kulakowski]

Is the idea to compile fast not under callgrind, stash the native code, and have a benchmark use that stashed version?