contours(): return_type validated too late; all-non-finite path silently accepts bad values

[brendancol]

Describe the bug

contours() validates return_type too late. The only check is at the end of the
function (xrspatial/contour.py:691), inside the dispatch chain:

if return_type == "numpy":
    return results
elif return_type == "geopandas":
    ...
else:
    raise ValueError(...)

That runs only after the raster has been validated, the levels computed, and the
contours extracted. Two problems fall out of the ordering:

1. A bad return_type isn't rejected until the whole extraction has already run, so
   the caller pays for work that was always going to fail.

2. The all-non-finite early return at xrspatial/contour.py:653-656 short-circuits
   before that check ever happens:

    if not np.isfinite(vmin) or not np.isfinite(vmax):
        if return_type == "numpy":
            return []
        return _to_geopandas([], crs=agg.attrs.get('crs', None))
   

   Any non-"numpy" value falls through to the GeoDataFrame branch. So
   contours(all_nan_raster, return_type="bad") hands back an empty GeoDataFrame
   instead of raising. The bad argument is silently accepted.

Expected behavior

return_type should be checked near the top of contours(), right after
_validate_raster and before any level computation or data work. A bad value should
raise ValueError no matter what the input raster looks like, including the
all-non-finite case.

Additional context

Fix: add an explicit return_type check (allowed values 'numpy' and 'geopandas')
at the top of the function. The existing late dispatch stays put but becomes
unreachable for bad values.