OpenGL `Transform`ing part of a `(Math)Tex` causes unexpected behavior relative to Cairo

Nobody has claimed this yet.

Consider the following scene:
```python
class Test(Scene):
def construct(self):
t1 = Tex("b")
t2 = Tex("a", "b")
t1.scale(5)
t2.scale(5)
t1.move_to(t2[1].get_center())
self.add(t1)
self.wait()
self.play(Transform(t1, t2[1]))
self.wait()
```
When rendered with Cairo, it works as expected: the b just sits there unmoving the whole time. However, when rendered with OpenGL, this happens:

https://user-images.githubusercontent.com/20936595/135917297-baf2526e-081b-478b-b8e7-6d7de11044de.mp4

Somehow manim doesn't recognize that the b in each case is the same and one of them shrinks to nothing while the other one grows from nothing. This is not the desired behavior when trying to create clean animations combining parts of equations.

Start by running the provided Test scene with both Cairo and OpenGL, focusing on Tex, Transform, and the two matching b glyphs. Compare how each renderer identifies and interpolates the corresponding parts; done means the OpenGL rendering keeps the b stationary, matching Cairo's behavior, while preserving the intended equation animation.

Written by the indexing model from the issue text.