Fix nested MatrixSpace scalar actions and coercions

[cxzhong]

This PR fixes the concrete nested-MatrixSpace action and coercion bugs identified while investigating #16247. It intentionally leaves the broader Modules(R) category-design issue open.

Root cause

When both operands belonged to matrix spaces, MatrixSpace._get_action_ selected matrix multiplication before considering that one matrix space was the other's base ring. Square nested matrix spaces also let the generic matrix constructor interpret a base-ring matrix as outer entries instead of using the canonical scalar embedding.

Changes

• Prefer left or right base-ring scalar actions over MatrixMatrixAction when one matrix space canonically coerces into the other's base ring.
• Embed a matrix base-ring element diagonally by multiplication with the outer identity, including for matrix semirings.
• For square targets, prefer the canonical S -> base ring -> target coercion path for matrix spaces and other matrix-like parents such as matrix groups.
• Construct an_element() from coordinate dictionaries when entries are matrices, avoiding the flat-versus-nested sequence ambiguity.
• Add regression tests for both action-discovery orientations, noncommutative multiplication order, coercible matrix subrings, dense and sparse spaces, dimensions 0 through 3, semirings, matrix groups, and ordinary matrix multiplication.

Validation

src/bin/sage-runtests --force-lib \
    src/sage/categories/modules.py \
    src/sage/categories/left_modules.py \
    src/sage/categories/right_modules.py \
    src/sage/categories/bimodules.py \
    src/sage/categories/action.pyx \
    src/sage/matrix/action.pyx \
    src/sage/matrix/matrix0.pyx \
    src/sage/matrix/matrix_space.py

All 1839 doctests passed. Nested parent and element TestSuite checks also passed after excluding the existing unsupported _test_minpoly check for matrices over a non-domain matrix ring.

📝 Checklist

• The title is concise and informative.
• The description explains in detail what this PR is about.
• I have linked a relevant issue or discussion.
• I have created tests covering the changes.
• I have updated the documentation and checked the documentation preview.

⌛ Dependencies

None.

[github-actions]

Documentation preview for this PR (built with commit 269f2e7; changes) is ready! 🎉
This preview will update shortly after each push to this PR.

[vincentmacri]

I imagine you're inferred this but just to be explicit: if I remove a request for review from me that just means I don't have the capacity for it at the moment and so you should try to request someone else. Please continue requesting reviews from me if you think it is a relevant part of the code that I can review, but I need to prioritize based on the amount of work and my own interest in the PR. I like to make sure I have enough time for when PRs that are especially interesting to me come in, like your recent set_to_product PR for matrix multiplication.