Reduce instruction fetch overhead

[explodingcamera]

Right now, wasm modules are stored in an Rc (as they are stored alongside host "modules), storing the instructions separately should bring a lot of speed improvements and maybe improve cache prefetching (?). The fastest way would be raw pointers, but I'd like to keep the codebase unsafe free.

Update: preliminary tests showed no tangible perf improvements from accessing instructions without indirection.
Update 2: bound-checking instructions adds around a 5% overehead in initial testing