Make sparse rational echelon forms immutable

[cxzhong]

Fixes #42531.

Problem

Matrix_rational_sparse.echelon_form() cached and returned a mutable matrix. A caller could mutate that shared result and corrupt every later cache hit, leaving the cached echelon matrix inconsistent with the source matrix's cached pivots and rank.

The already-echelon fast path also returned a mutable self. In addition, the multimodular helper returns self for matrices with zero rows or columns, so freezing its result directly would unexpectedly freeze the caller's matrix.

Changes

• Make computed sparse rational echelon forms immutable before caching them.
• Return and cache an immutable copy when a mutable matrix is already in echelon form.
• Copy zero-row and zero-column helper aliases before changing mutability.
• Document the immutable return contract and add regression doctests for all three paths.

Validation

• ./sage -t --long src/sage/matrix/matrix_rational_sparse.pyx (77 tests)
• Direct cache-corruption, pivot/rank, source-mutation, and empty-shape probes
• 42 randomized sparse shape and correctness checks against dense echelon forms
• git diff --check

[github-actions]

Documentation preview for this PR (built with commit b17e7a4; changes) is ready! 🎉
This preview will update shortly after each push to this PR.