Add `#[cold]` to `QueryIter::next_archetype`

[a1phyr]

Push further on the path opened by #336 by adding #[cold] to QueryIter::next_archetype

Results on my machine:

Before

test iterate_100k                    ... bench:      43,868 ns/iter (+/- 511)
test iterate_cached_100_by_50        ... bench:       1,003 ns/iter (+/- 13)
test iterate_mut_100k                ... bench:      44,154 ns/iter (+/- 601)
test iterate_mut_cached_100_by_50    ... bench:         142 ns/iter (+/- 42)
test iterate_mut_uncached_100_by_50  ... bench:         386 ns/iter (+/- 14)
test iterate_uncached_100_by_50      ... bench:       1,350 ns/iter (+/- 67)
test iterate_uncached_1_of_100_by_50 ... bench:         675 ns/iter (+/- 5)

After

test iterate_100k                    ... bench:      26,146 ns/iter (+/- 177)
test iterate_cached_100_by_50        ... bench:         595 ns/iter (+/- 9)
test iterate_mut_100k                ... bench:      26,362 ns/iter (+/- 363)
test iterate_mut_cached_100_by_50    ... bench:         141 ns/iter (+/- 3)
test iterate_mut_uncached_100_by_50  ... bench:         390 ns/iter (+/- 17)
test iterate_uncached_100_by_50      ... bench:       1,369 ns/iter (+/- 81)
test iterate_uncached_1_of_100_by_50 ... bench:         684 ns/iter (+/- 8)

[a1phyr]

Probably made obsolete by #342, but maybe you still want to keep that ?

[Ralith]

What's the before/after look like with #342? I recall not seeing any effect from this specific change on my machine, though variance between machines seems to be substantial for some reason.

[a1phyr]

With #342 this has no visible impact, this is why I am not sure if this should be kept ?