Note: `BTreeMap::remove<Q>` and `BTreeMap::remove<Q>` both accept a more general `Q` rather than just `Self::Key`.
If ever needed, `Map<Q>` would be possible (with the same bound on `Q` than `Self::Key`: `Ord` or `Eq + Hash`).

Implement `Map` for simple unordered `Vec<(_, _)>`. Note it uses `swap_remove`.

This is a huge but quite straightforward transformation:
- `GroupingMap[By]` becomes `GroupingGenericMap[By]` with a generic type `M`
- `K: Hash` becomes `M: Map<Key = K>`
- `HashMap<K, ...>` becomes `M` with a `M: Map<Value = ...>` bound
- Remove `Hash` and `HashMap`
- `use_std` becomes `use_alloc`

This is a breaking change! However, this is mostly inferred.
Note that it breaks (only) laziness tests because the `GroupingMap` object is not used (I could write any type but `u8` is one of the possible returned types). If it was used then it would have been inferred.

I copied/pasted the code of old `into_grouping_map[_by]` methods, added the generic `R` and updated each function body with `_in(.., HashMap::new())`.
The old types are aliases of the new ones.

rust-itertools#901 (comment)
SkiFire13 had the idea this could be optimized for vectors.