Refactor dialogue randomization

[bigkahuna443]

I separated out the randomized pools for spoken vs. nonspoken dialogue by sorting on "portrait" (can be made more specific with a regex but not really needed).

Also found a couple 3a bugs: -12-x (towels switch room): freezes if you enter part of the top hole before towels are set -00-d (presidential suite): if you retry after entering in the left (or some similar way of entering the room), the tweaked dash block will spawn, which softlocks you in 0-dash. You could also fix this by removing the dash block

[rhelmot]

so 1) you need to rebase, there are conflicts

but 2) there was a feature in the original implementation where it would try to pick a replacement dialog with approximately the same number of lines. I think something like that is worth keeping.

[bigkahuna443]

Oh my bad, looks like I didn't push my local merge... yeah I saw that offset method that they were using to "shuffle" the IDs... it had a set number of possibilities + could give you duplicates at one end of the list. I replaced it with a shuffle that breaks the list into segments and shuffles things within that segment, that should make things more random but still keep it within a similar length. Only issue is the current implementation will have an irregular sized segment at the very end, which means a few lines may have a smaller pool (but no duplicates). Not a huge deal IMO.

10 segments seems fine after a bit of testing, if people get too many default cutscenes that can be lowered to increase the pool for each segment.

I did have some fun combinations with randomizing without respect to length (e.g. the one I posted in the Discord) :) Can see pros/cons with either method

[bigkahuna443]

Going scorched earth on this one to fix Git issues :^)