async_std::sync::Mutex is too heavy

I've noticed that async-std uses `std::sync::Mutex` in quite a few places. It's well known that parking_lot offers much better performance, and more noticeably space saving, compared to the std mutex, thus enabling very fine-grained locks.

The current `async_std::sync::Mutex` and other few places of async_std seems to be using `std::sync::Mutex`, which has quite a few allocations and is very heavy-weight. Is there any existing discussion about performance and code bloat impact of switching over to parking_lot?

There is also a probability that we can use ideas similar to parking_lot to create a very cheap and light-weight (i.e. only a few bytes, no extra allocation) async Mutex that optimises for non-contended cases.