Optimize iteration

[Ralith]

Stumbled across some major speedups while chipping away at #334. Iteration of large archetypes is now ~twice as fast, with no significant change elsewhere.

Before:

test iterate_100k                   ... bench:     130,153 ns/iter (+/- 12,301)
test iterate_cached_100_by_50       ... bench:       1,010 ns/iter (+/- 75)
test iterate_mut_100k               ... bench:     127,847 ns/iter (+/- 6,311)
test iterate_mut_cached_100_by_50   ... bench:         251 ns/iter (+/- 58)
test iterate_mut_uncached_100_by_50 ... bench:         602 ns/iter (+/- 138)
test iterate_uncached_100_by_50     ... bench:       2,146 ns/iter (+/- 44)

After:

test iterate_100k                   ... bench:      51,880 ns/iter (+/- 2,813)
test iterate_cached_100_by_50       ... bench:         957 ns/iter (+/- 117)
test iterate_mut_100k               ... bench:      65,218 ns/iter (+/- 5,915)
test iterate_mut_cached_100_by_50   ... bench:         220 ns/iter (+/- 7)
test iterate_mut_uncached_100_by_50 ... bench:         701 ns/iter (+/- 548)
test iterate_uncached_100_by_50     ... bench:       2,340 ns/iter (+/- 205)

[Ralith]

Oddly this was much less impactful on my Ryzen 7800x3d desktop than my old i7-8550U laptop, though the outlining of next_archetype fixed iterate_mut being significantly slower than iterate there.