Chunk expensive numerical operations

[ValterH]

Requested by @ValterH: “create a new branch and properly merge 1016 into main.”

AI-generated draft summary: Bring #1016 onto main; the original PR was merged into mem-opt/base-numerical-chunks.

Add split_size and use it to compute RobustScale quantiles over column chunks and ClipSigma transforms over row chunks. Preserve the non-chunked autograd path and include chunk-equivalence checks.

Apply only the chunking commit on top of current main, retaining the final #1014 implementation without restoring its removed fitting-gradient test.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.