Refresh transform and nodata attrs on resample output

[brendancol]

Description

Three metadata gaps in xrspatial/resample.py found by the metadata propagation audit. These are the same issues PR #1462 fixed in xrspatial/reproject.

The end of resample() carries input attrs forward and overwrites res, but three other attrs stay stale:

1. transform describes the old grid. If the input had a rasterio 6-tuple transform, the output keeps the input value even though the cell size and origin have changed. Downstream code reading transform instead of res gets the wrong affine.

2. _FillValue keeps the input sentinel. Resample currently outputs float32 with NaN as the missing-data sentinel. If the input had _FillValue=-9999, the output still claims -9999 when no -9999 exists in the data: the missing pixels are NaN.

3. nodatavals (the rasterio plural form) keeps the same stale sentinel for the same reason.

Other attrs (crs, crs_wkt, scales, offsets, units, long_name) are correct unchanged because resample does not change CRS, datatype scaling, or units.

Proposal

At the end of resample():

• If 'transform' in agg.attrs, emit a fresh rasterio 6-tuple (res_x, 0.0, left, 0.0, -res_y, top) computed from the new edge coordinates.
• If '_FillValue' in agg.attrs, set _FillValue = float('nan') to match the actual sentinel in the output.
• If 'nodatavals' in agg.attrs, replace each entry with NaN.

Don't fabricate attrs that weren't on the input. Same gating approach as PR #1462.

Tests

• 4x4 raster with transform=(1.0, 0.0, 100.0, 0.0, -1.0, 200.0), scale_factor=0.5 produces output transform == (2.0, 0.0, 100.0, 0.0, -2.0, 200.0).
• _FillValue=-9999 on input produces output _FillValue of NaN.
• nodatavals=(-9999,) on input produces nodatavals=(nan,).
• Absent transform/_FillValue stays absent.

References

• PR Emit fresh grid metadata and propagate _FillValue through reproject() and merge() #1462 fixed the same issues in reproject() and merge().