Use a single Arc shared between Wakers in WakerArray and WakerVec

[wishawa]

This PR is dependent on #117

after = this PR
before = #117

group                after                                  before
-----                -----                                  ------
array::join 10       1.00      2.3±0.12µs        ? ?/sec    1.13      2.6±0.20µs        ? ?/sec
array::join 100      1.00     23.5±0.71µs        ? ?/sec    1.17     27.5±0.39µs        ? ?/sec
array::join 1000     1.10  1521.6±41.26µs        ? ?/sec    1.00  1377.7±21.95µs        ? ?/sec
array::merge 10      1.00      2.3±0.04µs        ? ?/sec    1.39      3.3±0.07µs        ? ?/sec
array::merge 100     1.00     52.5±6.18µs        ? ?/sec    1.09     57.5±2.20µs        ? ?/sec
array::merge 1000    1.00      2.5±0.06ms        ? ?/sec    1.04      2.6±0.07ms        ? ?/sec
array::race 10       1.00  1158.9±19.21ns        ? ?/sec    1.00  1159.8±25.91ns        ? ?/sec
array::race 100      1.00     12.2±0.33µs        ? ?/sec    1.02     12.5±0.48µs        ? ?/sec
array::race 1000     1.02    144.0±5.25µs        ? ?/sec    1.00    140.7±1.84µs        ? ?/sec
tuple::join 10       1.00      2.1±0.04µs        ? ?/sec    1.17      2.5±0.06µs        ? ?/sec
tuple::merge 10      1.00      2.8±0.06µs        ? ?/sec    1.10      3.1±0.07µs        ? ?/sec
tuple::race 10       1.05  1147.9±36.64ns        ? ?/sec    1.00  1089.3±19.36ns        ? ?/sec
vec::join 10         1.00      2.5±0.03µs        ? ?/sec    1.16      2.9±0.09µs        ? ?/sec
vec::join 100        1.00     34.5±1.06µs        ? ?/sec    1.15     39.6±0.80µs        ? ?/sec
vec::join 1000       1.07  1989.7±40.57µs        ? ?/sec    1.00  1863.9±26.66µs        ? ?/sec
vec::merge 10        1.00      2.9±0.13µs        ? ?/sec    1.25      3.6±0.06µs        ? ?/sec
vec::merge 100       1.00     57.3±1.04µs        ? ?/sec    1.28     73.3±1.38µs        ? ?/sec
vec::merge 1000      1.00      3.6±0.07ms        ? ?/sec    1.17      4.2±0.08ms        ? ?/sec
vec::race 10         1.01  1157.2±18.82ns        ? ?/sec    1.00  1147.6±14.78ns        ? ?/sec
vec::race 100        1.03     12.0±0.19µs        ? ?/sec    1.00     11.6±0.23µs        ? ?/sec
vec::race 1000       1.02    141.4±5.12µs        ? ?/sec    1.00    138.5±3.85µs        ? ?/sec

Also: made dummy waker easier to use and allocation-free.

[yoshuawuyts]

Thanks so much for this PR! - I'm trying to better understand everything that's going on, so this first round of reviews is mostly focused on documentation and clarifying what's going on in the unsafe blocks. I've mentioned this in the review comments as well: but the intent of this crate is to upstream parts of it to the stdlib, which means we need to build it with that in mind. I hope that makes sense; thanks!

[wishawa]

@yoshuawuyts thanks for the review! I added more documentation/comments, renamed things so that they (hopefully) make more sense, and simplified the code quite a bit[1]. Please take a look!

[1]
The simplification fixed the

if BY_REF {
            std::mem::forget(arc);
}

and

            wake::<A, false>,
            wake::<A, true>,

weirdness.

Previously I was using Arc::from_raw to get access to the Arc content, and dropping/forgetting the created Arc depending on whether the reference count should be decremented. In the updated code I dereference the raw pointer directly and, as you suggested, use Arc::decrement_strong_count properly.

[wishawa]

@yoshuawuyts @matheus-consoli is there still interest in this change? I think reducing the number of allocations from N to 1 is very beneficial.