geotiff: read_geotiff_dask does not support fsspec / cloud URIs

[brendancol]

Describe the bug

read_geotiff_dask() falls through to _read_geo_info() for non-HTTP sources, and _read_geo_info() uses plain open(source, 'rb') at xrspatial/geotiff/__init__.py:480. That call cannot resolve fsspec URIs (s3://, gs://, az://, memory://), so the dask path raises FileNotFoundError before any chunk task is built.

The eager path (open_geotiff -> _read_to_array -> _make_source in _reader.py) does route through _make_source and handles fsspec URIs via _CloudSource. The dask metadata read does not.

Reproduction

import fsspec, xrspatial.geotiff as xg

# Put a TIFF in fsspec's memory filesystem
fs = fsspec.filesystem('memory')
with open('local.tif', 'rb') as f:
    fs.pipe_file('/test.tif', f.read())

xg.open_geotiff('memory:///test.tif')               # works
xg.open_geotiff('memory:///test.tif', chunks=2)     # FileNotFoundError

Expected behavior

Dask reads should support the same source schemes as eager reads. _read_geo_info should route through _make_source / _is_fsspec_uri so cloud and memory URIs resolve identically.

References

• xrspatial/geotiff/__init__.py:1955 (dask path call site)
• xrspatial/geotiff/__init__.py:480 (offending open() call)
• xrspatial/geotiff/_reader.py:1020 (_make_source with fsspec branch)