InvertedPolarTransform also needs this?

Pushing to 3.8 as I do not think this is critical to get in and is still draft. Please push back if I am incorrect@dstansby .

We talked about this on the call this week and we felt we did not have enough context to understand why we want to do this and concern about exposing an underscored arg name to the users.

There was also surprise about the claim that this was not tested given it is the default!

Do we want to make this public if we are going to document it like this? Should we just have a second class?

I guess the point is that a polar axes sets this to False, which is our internal use. So this would only affect power users who are using PolarTransform directly? OTOH, it seems we should have a public toggle for this, and document it?

I think the original intention was to keep the 'version' of the class where no theta transform was applied private, but for backwards API compatability keep a version where a theta transform was applied. So I think making the argument public and going through a deprecation as is in this PR might be a reasonable way forward?

The motivation for this was to makePolarTransform internal code less complicated for maintenance reasons. I figured that there probably aren't lots of people out there using this, and there's a clear and simple alternative to_apply_theta_transforms=True (just add a theata value to the outputs), so it was worth doing.