Fix reset_transform in Cocoa/iOS

[HalfWhitt]

Fixes #4044

Adding the test first...

So far I have a fix that seems to work for Cocoa, but breaks iOS in new and interesting ways.

PR Checklist:

• I will abide by the BeeWare Code of Conduct
• I have read and have followed the CONTRIBUTING.md file
• This PR was generated or assisted using an AI tool

[HalfWhitt]

This and the test_attributes testbed test demonstrate that there are a lot of differences between backends in how they handle dashed lines at corners. It might be worth looking into whether / how specifically that's defined in the spec, and if so, if there are settings in the various platforms to conform more closely to it.

[HalfWhitt]

Interesting. Apparently while Cocoa uses a standard identity matrix when rendering to the screen and a different one for rendering to a cache, iOS never uses identity. I guess they handle flipping differently, somewhere along the way.

[freakboy3742]

The general approach here makes sense; a couple of questions about specifics of the implementation.

EDIT: Also - the comments about the Cocoa backend also apply to the iOS backend.

[freakboy3742]

I understand why we need to store these reference matrices - but do we ned to re-create them on every call to draw the widget? Won't they be the same on every usage?

[HalfWhitt]

At least on Cocoa, they'll be different if the call is to draw it to the screen, or to draw it to the buffer for as_image(). But that's a good point, we could just store each of those separately, as cached properties. And I think on iOS it's always the same — at least I know it's never the actual identity matrix. If I can figure out why Briefcase won't run iOS for me lately, I can go back in and double-check that.

[HalfWhitt]

Having refamiliarized myself with this a bit more: the original matrix (and its inverse) are dependent on the canvas's height. Granted, it should still be the same from draw to draw as long as the widget isn't resized, but we'd need to make sure to update it on resize. However, I've put the inverse inside a cached property, so it's only called if and when reset_transform is actually called (and still only once per draw).

[HalfWhitt]

(The test failure is from the Ubuntu domains being down again.)

[freakboy3742]

I think this all makes sense - thanks for getting to the bottom of all this.