pass a real rehash hasher to MapHashTable reserve / shrink

[sunshowers]

Previously, MapHashTable::reserve, try_reserve, shrink_to_fit, and shrink_to all
delegated to hashbrown::HashTable with |_| 0 as the rehash hasher.
When the capacity change triggers a reallocation, hashbrown's rehash
pipeline (resize_inner / reserve_rehash_inner) calls the supplied hasher
for every surviving entry and uses the result to pick its new bucket.
Returning 0 for every key crowds all entries into bucket 0 of the new
table; subsequent lookups probe from hash(key) mod capacity and miss,
so the map silently appears empty to get/find_index while iteration
still returns every item.

Also, reserve and try_reserve are reachable through the public API of
IdHashMap / BiHashMap / TriHashMap, including via Extend, which
pre-reserves size_hint / 2 before inserting.

The fix is to take a real hasher at the MapHashTable layer and plumb |ix| state.hash_one(items[*ix].keyN()) through every caller.

Also add coverage for these methods to our proptests.