Make rotate function compatible with cupy

The `dask_image.ndinterp.rotate` function currently supports numpy backed dask arrays, but not cupy GPU arrays. It would be nice to extend the functionality to add GPU support.

**Suggested approach:**
One way to approach this could be to:
1. Try and split out the sub-section of the `rotate` function that makes the rotation matrix, and put this into a separate helper function (i.e. group all the numpy specific lines of code together, while keeping the function as small as possible to reduce code duplication later).
2. Make a second version of the helper function using cupy/cupyx functions replacing all the lines that call numpy directly.
3. Then we could use the same dispatch mechanism as we do with the other modules (see the dask_image/dispatch folder) to register our new numpy and cupy versions.
4. Add a test for the new functionality
5. Update the `coverage.rst` table, adding a tick mark to indicate this function now has GPU support

Start with dask_image.ndinterp.rotate and inspect the existing dispatch implementations under dask_image/dispatch. Separate the rotation-matrix logic, provide NumPy and CuPy-backed paths through the established dispatch mechanism, and add a test demonstrating GPU-array support. Update coverage.rst so the rotate function is marked as GPU-supported.

Written by the indexing model from the issue text.