windowing should window along irregular CoordinateAxis

[cboulay]

Windowing on irregular data precludes the windows from being stacked, meaning that we can no longer guarantee that each iteration returns a simple AxisArray. We might need unique methods for these different circumstances.

The unique methods will be much easier to implement after the refactor from generator methods to class objects.

[cboulay]

One thing that will be tricky to handle is an incoming chunk that is longer than win_step. It's expected that an irregular-rate CoordinateAxis will have different numbers of samples per window, so we won't be able to yield a single AxisArray with multiple entries along the win axis.

Maybe it's enough to log a warning when the windowing generator func is configured in not 1:1 mode and the first incoming message's target axis is a CoordinateAxis. Then if we get a big data chunk that produces windows of different length then we raise an error.

I'd rather not support CoordinateAxis in non 1:1 mode because it will require yielding some container of multiple AxisArrays, which breaks compatibility with other processors.