Support 3D rasters, expose nodata, document target_resolution tuple in resample

[brendancol]

Reason or Problem

Three small API gaps in xrspatial.resample.resample() showed up while auditing API consistency against reproject():

1. 3D rasters are rejected. _validate_raster is called with the default ndim=2, so (band, y, x) inputs raise, even though reproject() accepts the same shape. Multi-band imagery has to be looped manually.
2. No nodata parameter. Integer rasters with sentinel values (e.g. -9999) cannot be resampled correctly without first converting to float and replacing the sentinel with NaN. reproject() already has a nodata keyword and falls back to _FillValue/nodata attrs.
3. The target_resolution docstring is wrong. The code at lines 801-803 already accepts a (res_y, res_x) tuple, but the docstring only documents float.

Proposal

Update resample() to:

1. Accept 3D rasters (band, y, x). Loop over the leading dim and stack the per-band 2D results. Output dims and non-spatial coords match the input.
2. Add a nodata=None keyword. When set, replace input pixels equal to nodata with NaN before resampling. When None, fall back to agg.attrs['_FillValue'], then agg.attrs['nodata']. Output uses NaN as the sentinel regardless.
3. Fix the target_resolution docstring to float or (float, float).

Value

Brings resample() in line with reproject() for multi-band and integer-sentinel inputs. No behavior change for existing 2D float callers.