Fix segfault when passing empty StringIO/File object to Font

[MyreMylar]

Related to #2295

Also fixes freetype submodule's Font to track submodule generation on __new__ instead of on __init__ for more reliable performance of the unit tests with uninitialized Font objects. Previously I was getting different results running on the CI versus locally.

[MyreMylar]

I also fixed two unrelated freetype module tests here that I noticed were failing locally while testing - but were not failing on the CI.

At some point a fix that was added for a quit() and init() segfault and the way the tests multithread on the CI causes these two particular tests which I changed, which use a class level __new__ constructor, to give one exception on the CI (that the module has been quit before the 'font' is created) and a different exception locally (the exceptions these tests were originally designed to capture).

It is not a real functional change but it means we can get these tests to pass both locally and on the CI.

[ankith26]

Thanks! Approving this PR because this LGTM, but I wonder if this is something that should also be fixed upstream on SDL_ttf if this is an issue on their end?

[ankith26]

Another improvement idea, we should use SDL_RWsize(source) instead of doing source->size(source). Shorter and also looks more robust

[MyreMylar]

Another improvement idea, we should use SDL_RWsize(source) instead of doing source->size(source). Shorter and also looks more robust

This function is available since SDL 2.0.10.
https://wiki.libsdl.org/SDL2/SDL_RWsize