Update `rustc_hash` and come to terms with the consequences

[sezna]

We appear to iterate over some hash maps in some places. Updating rustc_hash to a 2.x.x version causes all sorts of snapshot tests to fail with ordering issues.

I think we have two options:

1. Investigate these iterations and make them deterministic where important.
2. Update rustc_hash when we do a breaking change, so that our ordering can as well change.

Or...stay on 1.x.x forever?

[sezna]

@idavis has said to be sure that the performance doesn't degrade with this update -- they observed that in the past.