Summary

Pass-8 accuracy sweep finding. The JPEG-in-TIFF write path tags compression=7 (new-style JPEG) but writes a self-contained JFIF stream per tile/strip and never emits the JPEGTables tag (347) that the TIFF spec requires for that codec. libtiff, GDAL, and rasterio all reject the file with TIFFReadEncodedStrip() failed. Our reader round-trips because Pillow decodes the standalone JFIF bytes directly, which masks the break. The user thinks they wrote a TIFF; nothing else can read it.

Repro:
```python
import numpy as np, xarray as xr, rasterio
from xrspatial.geotiff import to_geotiff
arr = np.zeros((32, 32, 3), dtype=np.uint8); arr[..., 0] = 200
da = xr.DataArray(
arr, dims=['y','x','band'],
coords={'y': np.arange(32), 'x': np.arange(32), 'band': [0,1,2]},
attrs={'crs': 4326},
)
to_geotiff(da, '/tmp/x.tif', compression='jpeg', tiled=False)
rasterio.open('/tmp/x.tif').read() # CPLE_AppDefinedError: TIFFReadEncodedStrip() failed
```

Fix: reject compression='jpeg' at the to_geotiff entry point with a clear ValueError pointing at deflate/zstd/lzw. The lower-level _writer.write is untouched so the existing self-decoding tests still cover the codec until someone wires up a JPEGTables-aware encoder.

Test plan

• new test_to_geotiff_jpeg_rejected covers the user-facing error
• existing _writer.write JPEG round-trip tests pass unchanged
• full geotiff suite: 671 passed, 4 skipped (3 unrelated matplotlib deepcopy failures predate this change)