NOTE: CI failing at this point

Bug was caused by reusing the adjusted strides in the logic to decide whether the call to GEMV should be transposed or not.

Particularly the +1 in the strides variable was causing the error branch (no double-strides) to be reached wrongly. The +1 was supposedly there for the case of matrix with length 0, but that triggers a branch where the adjusted strides are never used.

This bug was introduced in afe934b2

Bumps [pypa/cibuildwheel](https://github.com/pypa/cibuildwheel) from 2.23.0 to 2.23.3.
- [Release notes](https://github.com/pypa/cibuildwheel/releases)
- [Changelog](https://github.com/pypa/cibuildwheel/blob/main/docs/changelog.md)
- [Commits](https://github.com/pypa/cibuildwheel/compare/v2.23.0...v2.23.3)

---
updated-dependencies:
- dependency-name: pypa/cibuildwheel
  dependency-version: 2.23.3
  dependency-type: direct:production
  update-type: version-update:semver-patch
...
Signed-off-by: dependabot[bot] <support@github.com>

Set view flag of ExtractDiagonal to True and respect by default

Split off Subtensor of Unbroadcast into its own rewrite

It seems to add a non-trivial 100ns

Signed-off-by: Emmanuel Ferdman <emmanuelferdman@gmail.com>

Also add checks for runtime broadcast

These show up in the gradient of Convolve1D