fast-jit performance issue

[yamt]

consider fast-jit with WASM_ENABLE_LAZY_JIT.
background compiler threads (orcjit_thread_callback) hold fast_jit_thread_locks most of the time until they finish their work.
on-demand compilation (compile_fast_jit_and_then_call) needs to compete with the corresponding background thread and win the race to make a progress.
depending on the mutex implementation and thread scheduling, the on-demand compilation can be starved.

possible solutions:
a. retire the background threads
b. make the lock finer grain (eg. per target function)
c. implement a fair lock and use it instead of the os mutex