`chunks` needs to be passed to `to_multiscale` otherwise it's ignored

Super easy fix.

Problem

The function to_multiscale(), which is called by Image2DModel and Image3DModel when scale_factors is a list, calls this code:

    # IPFS and visualization friendly default chunks
    if "z" in image.dims:
        default_chunks = 64
    else:
        default_chunks = 256
    default_chunks = {d: default_chunks for d in image.dims}
    if "t" in image.dims:
        default_chunks["t"] = 1
    out_chunks = chunks
    if out_chunks is None:
        out_chunks = default_chunks

This means that if some chunks were already set for the data passed to the model, as in this case

    # data = da.ones((3, 32768, 32768), chunks=(1, 4096, 4096))
    chunks= (1, 4096, 4096)
    data = RNG.random((3, 32768, 32768), chunks=chunks)
    xdata = DataArray(data, dims=("c", "y", "x"))

    ##
    im = Image2DModel.parse(
        xdata,
        scale_factors=[2, 2, 2],
        # chunks=chunks
    )

They are rechunked, unless we pass chunk explicitly to Image2DModel.parse().

Solution

If the data has already chunks, pass them to to_multiscale(). This needs to be done when the data with a Dask array or an xarray DataArray.

Start by tracing to_multiscale() and its calls from Image2DModel and Image3DModel when scale_factors is a list. Check the Dask-array and xarray.DataArray paths, then verify that existing chunks are passed through instead of replaced by defaults. Done means explicitly chunked input retains its chunks without requiring chunks on Image2DModel.parse().