feat(driver): move future state into RawOp

[Berrysoft]

In this way, no need the OpRuntime, and the HashMap could be removed.

[Berrysoft]

f6e7c4a adds a large patch to reduce the usage of slab in compio-driver. The allocated pointer itself is a good user-defined data for the operation. On Windows, it is also the pointer of OVERLAPPED struct.

However, the patch makes the code less readable. There are many conversions between pointer and usize, and the RawOp even constructs its own vtable to keep the pointer thin. The performance doesn't increase significantly.

@George-Miao @oxalica What do you think of the patch?

[George-Miao]

If I'm understanding correctly:

• RawOp has to be repc(C) and op has to be the last field so that Key<()> will have the same prefix as Key<T>
• Each upcast::<T> is used as an unique fn ptr to construct the correct dyn vtable from user data for T: OpCode

Is that correct?

[Berrysoft]

Bingo!