chunking pressure in eos at the wrong time?

[mnlevy1981]

@sgyeager ran into an issue where calling eos() when salt and temp are dask arrays but depth is a numpy array (well, all three are of type DataArray, but under the xarray wrapper it's dask, dask, numpy) took several minutes while still returning a dask object, but chunking depth so it was a dask array as well let eos() finish in under a second. I think the problem is in https://github.com/NCAR/pop-tools/blob/main/pop_tools/eos.py#L100

        salt, temp, pressure = xr.broadcast(salt, temp, pressure)
        if isinstance(salt.data, dask.array.Array):
            pressure = pressure.chunk(salt.chunks)

wouldn't it make more sense to chunk pressure before the xr.broadcast()? I.e.

        if isinstance(salt.data, dask.array.Array):
            pressure = pressure.chunk(salt.chunks)
        salt, temp, pressure = xr.broadcast(salt, temp, pressure)

[sgyeager]

It turns out that sending depth as a dask array causes a Failed to Serialize error, so this is not a fix. There appears to be issues with eos() when salt and temp are large dask arrays. It can take several minutes to complete the lazy eos() call and memory use on master can grow huge. I have not pinpointed the problem yet.