This better matches CC2 behavior and also makes some very common
operations, like grabbing a cell's actor or terrain, way faster.
It also allows me to efficiently implement CC2's layer order when
checking for collisions; thin walls are checked before terrain, and
actors only afterwards. The upshot is that bowling balls no longer
destroy stuff on the other side of a thin wall!
I also did some minor optimizing, mostly by turning loops over an entire
cell's contents into checks for a single layer; Chromium now performs a
bulk test about 30% faster.
Downsides of this change:
- All kinds of stuff may have broken!
- It'll be a little difficult to ever emulate MSCC's curious behavior
when stacking terrain on top of items or other terrain. But not
impossible.
- It'll be far more difficult to emulate buggy Lynx (or maybe it's just
Tile World?) behavior where some combination of cloners and teleports
allow a ton of monsters to accumulate in a few cells. I guess I
wasn't planning on doing that anyway.
Mostly style nits, but also:
- Renamed some stuff in anticipation of removing GameEnded.
- Actor decisions are independent, so there's no need to do most of them
in the first part of a tic and the player in the second part; they can
all happen together in the second part.
- waiting_for_input was merged into turn_based, which I think makes it
easier to follow what's going on between tics. Although I just
realized it introduces a bug, so, better fix that next.
- The canvas didn't need to know if we were waiting or not if we just
force the tic offset to 1 while waiting. This also fixed some slight
jitter with force floors.
