Good that you got a repro now at least! You do have a grid as a model where you either set the grid occupied or not and then moving the visual is built on top of the grid? I think that way it should be solvable quite easily. But then of course it means that you need to adjust the player movement against the grid if player can move freely. And still, in the grid solution if two neighboring cells are occupied the player can't get between and game is stuck. :)
Viewing post in Paco-Cabana jam comments
The player should (at least in theory) also just move in grid cells. When using a gamepad's analogue stick, when movement is registered it's normalized for a grid cell's length.
But the way I implemented it, I rely too much on mapping towards the tilesheet grid, so I may get some inaccuracies with detecting neighbouring cells and what state they're in. I should completely rewrite all that stuff to work in a proper grid array...