Reduce type: ignore count in _config/utils.py and extract fit-to-size logic from mobject.py

[viniciuspalmieri]

Draft PR addressing the two cleanups discussed in #4734 and #4735. Marking as draft to signal that work is in progress and to invite feedback on the approach before requesting review.

Closes #4734
Closes #4735

Commits

This PR contains two independent commits, one per issue, so either can be reverted without affecting the other.

1. Reduce # type: ignore count in _config/utils.py (#4734)

Brings the file from 10 markers down to 4. The six removed all came from annotations being slightly looser than reality rather than the underlying code being genuinely hard to type:

• Tightening _d: dict[str, Any | None] to dict[str, Any] removes two [operator] ignores at the frame_y_radius / frame_x_radius getters.
• Correcting __deepcopy__'s memo annotation from dict[str, Any] to dict[int, Any] removes one [arg-type].
• Rewriting the a.__setitem__(...) or b.__setitem__(...) chain hack in three setters as two plain assignments removes three [func-returns-value] ignores.

The four remaining markers (Liskov in update(), has-type bootstrap on _tex_template, literal-required on a QUALITIES key, dynamic setattr on a lambda) are structurally harder and left out of scope.

2. Extract fit-to-size logic from mobject.py (#4735)

Moves the implementation of rescale_to_fit and the six *_to_fit_* wrappers out of Mobject and into a new manim/mobject/_fit.py module of free functions. The seven methods remain on Mobject as thin stubs that delegate to the utility functions.

Shape (utility + delegation rather than a mixin) follows the suggestion from #4735 and matches the established pattern in manim/utils/ (space_ops, iterables, etc.). It also avoids the # type: ignore markers a mixin would need, since the utility functions take the mobject as a regular parameter.

The public API is unchanged: Square().scale_to_fit_width(5) works identically. mobject.py's overall LOC count is roughly unchanged (the stubs and docstrings stay), but the substantive implementation now lives in _fit.py (51 LOC). This is meant as a precedent for further cohesive groups (transforms, positioning, color), without claiming a position on a broader split.

Validation

Run locally on Windows 11 / Python 3.12 / uv 0.9.29:

• ruff check clean on touched files.
• mypy clean on manim/_config/utils.py and manim/mobject/_fit.py.
• pytest tests/module/mobject/mobject/ tests/test_config.py: 30 passed, 0 failed (3 tests deselected because they fail identically on unmodified main due to missing latex/dvisvgm/manim CLI on the local machine; verified via git stash).
• Extended pytest on non-LaTeX-dependent modules: 138 passed.
• Targeted: pytest -k "scale_to_fit or stretch_to_fit or frame_y_radius or frame_x_radius or frame_size": 3/3 passed.

I haven't run the graphical_units suite (requires ffmpeg); given the nature of the changes (no rendering logic touched), I don't expect issues, but flagging it.

Open questions before moving out of draft

1. For Reduce # type: ignore count in manim/_config/utils.py #4734: would you prefer this as a single PR for all six fixes (as here), or split into three (one per category)? The categories are small enough that one PR feels right to me, but splitting makes pieces independently revertable.
2. For Extract fit-to-size methods from mobject.py into a mixin module #4735: confirming the utility + delegation shape is preferred over the mixin shape I initially proposed.
3. Any objection to the two cleanups landing together, or would you rather have them as separate PRs?