Leverage dask directly to improve compute_MVBS efficiency

[leewujung]

During investigation of #834, I found an efficiency issue in the current implementation of compute_MVBS if the data is delayed. For the 19 files in the OOI eclipse notebook, if the ds_Sv data are persisted in memory it takes 5 mins to compute, but if ds_Sv is lazy-loaded the operations never finished.

Further investigation by @b-reyes showed that the xarray groupby_bins operations are the major hiccups. The under-the-hood implementation can likely be improved dramatically by doing indexing directly using dask, and we can do this in the next release.

[leewujung]

Temporary fix? @lsetiawan will try to see if we can delay the function for now to allow this to run through.

the lines that use xarray functions to do mean:

echopype/echopype/preprocess/api.py

Lines 57 to 62 in c4a021b

	sv_groupby_bins = (
	sv.resample(ping_time=pbin, skipna=True)
	.mean(skipna=True)
	.groupby_bins("echo_range", bins=rint, right=False, include_lowest=True)
	.mean(skipna=True)
	)

[leewujung]

This is now addressed in #878.