separate compilation process

[thautwarm]

After generating cython code, the compilation(cythonize + cc compile .c to .pyd) could be proceeded in a separate process.

Without this, there will be a notable JIT latency.

root_func = jit_root(func)
root_func(arg1, arg2) # root_func uses the original implementation if func hasn't been specialised and compiled yet.

[thautwarm]

No need so far. JITing a method now is damn fast. However, I'd keep this issue, as @lfkdsk pointed out, it is possible for python to send compilation tasks to multiple processes, which will send back the tasks to julia, which is in the same process with python.