to_geotiff has no way to set LERC max_z_error

[brendancol]

Problem

to_geotiff(..., compression='lerc') always calls lerc_compress with max_z_error=0.0 (lossless). LERC's main draw is that you can trade a small per-pixel error budget for big size savings on float data, and there is currently no public way to set that budget.

xrspatial/geotiff/_writer.py:401-404, 464-467, 1106-1109 all call lerc_compress(strip_data, ..., samples=samples, dtype=dtype) without forwarding a max_z_error argument, so it falls through to the function default of 0.0 in _compression.py:1018-1020.

Repro

```python
import numpy as np, xarray as xr, os
from xrspatial.geotiff import to_geotiff

data = xr.DataArray(np.random.rand(1024, 1024).astype(np.float32))
to_geotiff(data, '/tmp/lossless.tif', compression='lerc')
to_geotiff(data, '/tmp/lossy.tif', compression='lerc') # no way to ask for lossy

print(os.path.getsize('/tmp/lossless.tif'),
os.path.getsize('/tmp/lossy.tif')) # identical
```

Suggested fix

Add max_z_error: float = 0.0 to to_geotiff and forward it through _write_stripped, _prepare_tile, the streaming writer, COG, and VRT paths into lerc_compress. Validate that max_z_error == 0 (or raise) when the selected compression is not LERC, so callers don't silently get the value ignored.