Use new iterator

[davidamarquis]

This PR implements a multithreaded version of Brouwer's algorithm. In the old approach a matrix multiplication was done for every iteration. The new approach eliminates all matrix multiplications except for a single initial one. This is done by using Gray code iteration with a revolving door algorithm over the messages of a given weight.

Benchmarks

All benchmarks use 3 threads. These codes are used by White in his thesis.

A [70, 35, 14] code

Old Approach
Range (min … max): 3.072 s … 3.283 s ┊ GC (min … max): 0.00% … 0.00%
Time (median): 3.178 s ┊ GC (median): 0.00%
Time (mean ± σ): 3.178 s ± 149.711 ms ┊ GC (mean ± σ): 0.00% ± 0.00%
Memory estimate: 70.49 MiB, allocs estimate: 4583915.

This PR
Range (min … max): 148.035 ms … 157.587 ms ┊ GC (min … max): 0.00% … 0.00%
Time (median): 148.730 ms ┊ GC (median): 0.00%
Time (mean ± σ): 149.059 ms ± 1.599 ms ┊ GC (mean ± σ): 0.00% ± 0.00%
Memory estimate: 543.20 KiB, allocs estimate: 1434.

A [105, 35, 26] code

I think this example will enumerate about 10X more vectors than the [70, 35, 14] one (about 50 million vectors).

Old Approach
I killed the run after it did not finish in 30 minutes.

This PR
Single result which took 44.510 s (0.00% GC) to evaluate,
with a memory estimate of 1.12 MiB, over 2532 allocations.