[ENH] twe memory and speed improvements

[TonyBagnall]

Part of #3261 (memory optimization for elastic distances). Applies the two-row dynamic-programming optimization established for DTW in #3258 (and MSM in #3562) to TWE.

closes #3269

• Public API and outputs are unchanged. Signatures, defaults (nu=0.001, lmbda=1.0), docstrings, and return values of twe_distance, twe_pairwise_distance, twe_cost_matrix, and twe_alignment_path are all the same.
• The full cost matrix is intentionally retained. _twe_cost_matrix is untouched and still backs twe_cost_matrix (which by definition returns the whole matrix) and twe_alignment_path (which needs the full matrix to backtrack the alignment via
  compute_min_return_path).
• _pad_arrs, the pointwise Euclidean cost, the pairwise / multiple-to-multiple drivers, and the bounding-matrix / windowing / Itakura logic are untouched. All callers of _twe_distance (the pairwise functions) already pass padded arrays and are
  unaffected.

Validation

• Equivalence: matches the previous full-matrix result to floating-point tolerance — max absolute difference ~3.5e-14 over 400 randomized cases (univariate/multivariate, varying nu/lmbda, windows, and unequal lengths), growing to ~1e-9 at n=8000
  (≈1e-13 relative).

Benchmark (univariate, n = m, no window)

n	Runtime baseline → opt	Peak mem baseline → opt	Mem reduction
1,000	15.6 ms → 13.6 ms (1.1×)	9.0 MB → 1.0 MB	9.0×
2,000	67.9 ms → 60.1 ms (1.1×)	36.0 MB → 4.0 MB	9.0×
4,000	0.26 s → 0.21 s (1.3×)	144.1 MB → 16.1 MB	9.0×
8,000	1.89 s → 0.85 s (2.2×)	576.3 MB → 64.3 MB	9.0×

The optimization eliminates the float64 cost matrix (n²·8 bytes); the ~9× reduction reflects that the remaining memory of the public call is the O(n²) boolean bounding matrix (n²·1 byte, window=None).

[aeon-actions-bot]

Thank you for contributing to aeon

I have added the following labels to this PR based on the title: [ enhancement ].
I have added the following labels to this PR based on the changes made: [ distances ]. Feel free to change these if they do not properly represent the PR.

The Checks tab will show the status of our automated tests. You can click on individual test runs in the tab or "Details" in the panel below to see more information if there is a failure.

If our pre-commit code quality check fails, please run pre-commit locally and push the fixes to your PR branch.

Don't hesitate to ask questions on the aeon Discord channel if you have any.

PR CI actions

These checkboxes will add labels to enable or disable CI functionality for this PR. This may not take effect immediately, and a new commit may be required to run the new configuration.

• Run pre-commit checks for all files
• Run mypy typecheck tests
• Run all pytest tests and configurations
• Run all notebook example tests
• Run numba-disabled codecov tests
• Disable numba cache loading
• Regenerate expected results for testing
• Push an empty commit to re-run CI checks

[TonyBagnall]

[TonyBagnall]

[TonyBagnall]

speed up less here, but memory gain the same