Review block_ends in block_analysis.Block

[jkbhagatio]

Right now we're just detecting block ends by seeing if there's a negative diff in pellet_ct values in the block_state file. However, this can fail if there are consecutive blocks with 0 pellets obtained.

Instead, we can either do:

1. look for 0's in the block_state file, and if there is a large enough gap between consecutive 0's, assume the block ended.

or

2. use the NaN values in PatchState that occur on block transition.

3. is probably preferred, since it uses the block_state info directly