SWY: check for numpy.nan pixels in routing functions

[davemfish]

We have updated our is_close and numpy.isclose calls in most places to check for numpy.nan, but this is one place where we were not doing that. It's plausible for the rasters being processed by this module to have numpy.nan pixels.

Fixes #1705

Checklist

• Updated HISTORY.rst and link to any relevant issue (if these changes are user-facing)
• Updated the user's guide (if needed)
• Tested the Workbench UI (if relevant)

[emlys]

Thanks @davemfish! Just wondering if this has an effect of slowing down the model significantly?

[davemfish]

Thanks @davemfish! Just wondering if this has an effect of slowing down the model significantly?

@emlys I did some tests and found no noticeable slowdown of the model until I used a datastack that took 56 minutes to run on my machine. In that case, a version with the isnan checks took 57 minutes.

At a recent meeting we discussed this and agreed this is acceptable for now since doing the isnan checks fixes a bug where nans would otherwise pollute some math.

I also created #1714 and #1715 so that we can revisit a broader invest-wide solution and revert this change later if we want to.

[emlys]

@davemfish thanks, I was going to wait for the checks to pass, but it seems that the macos checks have been stalled all day (on this and other branches too), so feel free to merge given that all seemingly relevant checks have passed.