The Witness: New hint type ("area hints")

[NewSoupVi]

Adds a new option called "Area Hint Percentage".

So far, Witness hints refer to singular locations, i.e. "Shipwreck Vault Box contains Hookshot (Player 4)"

This new type of "area hint" will instead give you general information about one of the named geographical areas in your world.
Example:

There are 4 progression items in the "Quarry" region.
Of them, 2 are for other players.
Also, one of them is a laser for this world.

This also renames some of the locations in the game to better fit into an "area", such as the "River Obelisk" being renamed to the "Mountainside Obelisk".
For this reason, it will block future PRs such as #2805 because I don't want a bunch of "soft conflicts" from location name mismatches.

Also added location groups based on these new "areas".

Testing

100+ random settings test generations.
Many generations where I breakpoint after all hints are created and look at all of them.
Several full seeds played.

Client

There is a compatible client in prerelease.

Special handling if you're curious

1. Not enough areas

There are only ~20 areas in the game but you can have 49 hints with an area_hint_percentage up to 100%. The generator will generate as many area hints as possible and then make random location/item based hints.

But, this creates another strange situation:
Say the player wants 49 hints with 100% area hints.
They would now instead get 20 area hints and 29 location/item hints.
However, those 29 hints would be fully random rather than acting like normally generated hints. This could potentially be confusing in mystery settings
To prevent this, the generator keeps track of "always hints" and "priority hints" that it did not use, and tries to "reuse" them if possible.

2. "Duplicate hints"

A lot went into not having it happen that an area is hinted, and then all locations in it are seperately hinted. It'll only ever hint all but one location individually in an area-hinted area.

[Exempt-Medic]

Merged regional_hints into main, did 1000 generations (no errors), read through several spoiler logs and generated hint lists. Played at least 20 runs with the changes. All the text and location changes LGTM. Specifically tested the weighted_sample code quite a lot. Read through the rest of the code as well, which LGTM aside from some comments that are now resolved.