ROI support in LiveContext.run_udf

[matbryan52]

Currently we have no support for using a nav-space ROI during live processing, and the behaviour is undocumented / implicitly allowed via the normal signature of run_. At least in my testing using a non-full ROI results in PipelineExecutor 'dropped tasks' or incorrect number of frames-type errors. For the moment we could add a warning (or even an exception) if roi is not None or contains zeros, on LiveContext.

That said, if we can find a way to handle an ROI there might be benefits in performance, as we're not required to buffer or process all frames, though speaking to @sk1p this isn't a guarantee.

One real use case for ROIs are cameras for which the data stream contains 'bad' frames, like a column of flyback frames or a line at the end of the scan which should be dropped (needed for trigger timing reasons). In particular for Ptychography having bad frames like that could negatively impact the reconstruction. On a more basic level it can also throw off plot colormap scaling.