Avoid copying vectorized indexes

[jder]

While investigating #10311, I noticed that the large memory usage described there was exacerbated by duplicating the passed indexes in VectorizedIndexer.__init__. While this could be a defensive copy to avoid someone else mutating the passed ndarrays, looking at the PR it was introduced I would guess this was an unintentional regression as part of supporting duck arrays, since asarray defaults to not copying the data and astype defaults to copying the data, but cc @dcherian in case that's not true.

Also included in this PR is increasing the size of the benchmark to show the improvement more clearly. (On my machine, it's a ~20% time savings.) Happy to break this out or remove it if you'd prefer to leave the benchmarks as is.

• Tests added

[welcome]

Thank you for opening this pull request! It may take us a few days to respond here, so thank you for being patient.
If you have questions, some answers may be found in our contributing guidelines.

[dcherian]

Very nice! I have two minor requests:

1. Can you use duck_array_ops.astype instead?
2. There is another astype in OuterIndexer that could be updated.

Can you also add a note in whats-new.rst under "Performance" please

[jder]

OK, @dcherian I believe this is RFAL. The test failures seem to be because dask released a new version yesterday -- for the ones I spot checked, I can reproduce them locally with dask==2025.5.0 but not dask==2025.04.1.

[welcome]

Congratulations on completing your first pull request! Welcome to Xarray! We are proud of you, and hope to see you again!

[dcherian]

Thanks for tracking this down @jder! I know this is not an easy part of the codebase to edit :)

[jder]

@dcherian Thanks for the quick review, feedback, and landing it!

[alxmrs]

🎉