FIX: use shoelace formula for input polygons

[rcomer]

Rationale

Following #2665, also use the more robust shoelace formula when checking the direction of polygons before we transform them.

Fixes #2312. In that example, when projecting the data before calling contourf we get

If we call contourf on the original data, with main we get

and with this branch we get

I wrote this branch back in April and saw 179 test failures. Those are now gone, but I have not tracked down why. I checked what happened if I apply the change to the commit just before #2653, but that also passed. Edit: it's possible I was running with proj 9.8 so was mistaken that the branch caused the failures.

Implications

[rcomer]

I do not understand the MacOS failure, but I also do not think it is related to this change as it has to do with creating a CRS object.

[greglucas]

It looks like there is another one in patch.py, should we apply it there too?

[rcomer]

I'm inclined to leave the one in patch.py alone: the function is deprecated and there appear to be no tests covering force_ccw=True, which is the case when the check is used. It would almost certainly do no harm to change it, but the lack of coverage makes me twitchy.

[greglucas]

Ahh yes, I forgot that was deprecated. Sounds good!