inline poll_states and remove Fuse for vec::merge

[yoshuawuyts]

Ref #22

Counterpart to #70. Removes the use of Fuse from impl Merge for Vec, and also inlines the poll_states.

Performance

Compared to main
Performance remains roughly comparable.

❯ critcmp main patch  -f vec::merge
group              main                                   patch
-----              ----                                   -----
vec::merge 10      1.04      2.7±0.12µs        ? ?/sec    1.00      2.6±0.05µs        ? ?/sec
vec::merge 100     1.00     43.2±1.93µs        ? ?/sec    1.10     47.5±0.97µs        ? ?/sec
vec::merge 1000    1.02      2.0±0.19ms        ? ?/sec    1.00      2.0±0.15ms        ? ?/sec

Compared to last release

The delta here remains mostly dominated by the move to "perfect" waker semantics:

❯ critcmp patch baseline  -f vec::merge
group              baseline                               patch
-----              --------                               -----
vec::merge 10      1.93      4.9±0.31µs        ? ?/sec    1.00      2.6±0.05µs        ? ?/sec
vec::merge 100     7.60    360.6±3.84µs        ? ?/sec    1.00     47.5±0.97µs        ? ?/sec
vec::merge 1000    18.94    38.0±0.41ms        ? ?/sec    1.00      2.0±0.15ms        ? ?/sec

[eholk]

Looks good! I had a couple small suggestions for potential improvements, but I'll leave them up to you if you want to take them. Feel free to merge whenever!

[eholk]

I assume this is the number of streams we're merging? I might suggest count or num_streams as alternate names, but there's plenty of precedent for using len for this sort of thing, so I'm happy to leave it up to your judgement.

[eholk]

Actually, I'd suggest adding a len() method that returns self.streams.len() and that way we save 8 bytes off the size of the Merge struct.

[yoshuawuyts]

Ah, I wish we could set methods! - Because we're dealing with a pin projection it generates a struct for us containing the projection. Andy any self methods would need to point to that.

Because we don't control the projection we can't define methods. And so we basically have to define everything in-line. Which is like, another reason why I'd reallllly like us to have safe pin projections, so we can split manually authored futures into separate internal methods etc. Not being able to do this is another reason authoring futures by hand is hard :/