Review and improve HashMap's handling of an empty table

[RalfJung]

In #53804, I fixed the most immediate problem arising from how HashMap handles the empty table, but @gnzlbg still (rightly) sees more problems. We still use 1 instead of NonNull::dangling() as value for EMPTY, and ptr() even tests equality with EMPTY to return NULL instead. We can't really use non-NULL pointers as sentinel values like that, and NULL is often a worse choice than an aligned dangling pointer because the latter are at least valid for zero-sized operations.

See here for @gnzlbg's suggestions.

[Mark-Simulacrum]

I'm going to close this issue since our hash maps are now a thin wrapper around hashbrown which hopefully doesn't suffer from these issues; if it does, this should be refiled upstream. A cursory examination didn't find any obvious problems for sentinel use of dangling or so.

[gnzlbg]

Yes none of this applies to the new hashbrown impl. cc @Amanieu in case they want to re-read the comments and any of that rings a bell, but I haven't see anything like that in hashbrown.

[Amanieu]

I agree, I don't think any of this applies to hashbrown.