Sort preferences by environment, then index

[charliermarsh]

Summary

This has a few effects:

1. We only call preferences once, which should be more efficient.
2. We collect preferences into a vector when there are multiple. Less efficient, but pretty rare?
3. We now correctly prefer preferences from the same index.

[konstin]

We can use a smallvec here to avoid the allocation for diverging forks (with 2 elements it's even the same size, though this is stack only anyway)