Harmony 992

[hailiangzhang]

This PR will introduce rechucking during netcdf-to-zarr step.

The destination chunk size will be the multiplier of the original chunk size closest to 3000, or 3000 if the original chunk size is more than that.

If I don't explain it well, the formula is here from convert.py:

116     new_chunks = map(
117         lambda x: min(x[0], int(3000 / x[1]) * x[1] if x[1] < 3000 else 3000),
118         zip(shape, chunks),
119     )

Basically the idea is to make the destination chunk size to be "aligned" with the original chunk size (for performance reason), and 3000 is a number to avoid memory issue with the current netcdf-to-zarr container size.

We may not see too much performance improvement here, which will be the focus of the subsequent related stories.