Document that block splitter doesn't know about padding when reading blocks

In the task runner, the max slices is calculated and then reduced by the number of padding slices, before being given to the block splitter: https://github.com/DiamondLightSource/httomo/blob/fbfbc2032c3005c69f354e834012bb064dc3c3ce/httomo/runner/task_runner.py#L106-L111

The block splitter in turn will read blocks from its `DataSetSource` implementor, unaware of the blocks being read will potentially contain padded slices. (Ie, if padding is `(2, 2)` and the block splitter is told to read a block of length 5, it'll get a block with slicing dim length 5 + 2 + 2 = 9).

It'd be useful to document this more clearly _somewhere,_ before I forget this fact once again...