Deal with hidden dask usage

[dfulu]

We have moved to primarily use xarray-tensorstore as a backend, but in some places it is possible for dask to creep back in. I recently had this where there was an extra unused non-dimensional coordinate in our ECMWF data. This coordinate wa maintained up to the compute method and since it uses dask, even though the coordinate is small and has very few chunks, it slowed down my sampler by around 20-30%.

This is something that could keep coming up.

We should perhaps:

• Remove non-dimensional coordinates in the load functions?
• Add a warning to the user if any datasets are dask arrays

[Joy-chakraborty-23]

@peterdudfield and @dfulu would it be okk if i you guys assign me this issue ?

[peterdudfield]

Thanks @Joy-chakraborty-23, I think @dfulu is tackling this.

I might be good to look out for good-first-issues or contributors-welcome labels, as they are often tasks more suitable