read_vrt: SrcRect/DstRect scaling not applied to source array placement

[brendancol]

read_vrt does not resample source pixel data when a source band's <SrcRect> size differs from its <DstRect> size. Any VRT that uses GDAL's source-to-destination scaling either raises ValueError (on downsample) or returns spatially wrong pixels with the rest of the destination cell left at the fill value (on upsample).

Reproducer:

import numpy as np, tempfile, os
from xrspatial.geotiff._writer import write
from xrspatial.geotiff._vrt import read_vrt

with tempfile.TemporaryDirectory() as tmp:
    arr = np.arange(16, dtype=np.uint16).reshape(4, 4)
    src_path = os.path.join(tmp, 'src.tif')
    write(arr, src_path, compression='none', tiled=False)

    vrt_path = os.path.join(tmp, 'down.vrt')
    with open(vrt_path, 'w') as f:
        f.write(f'''<VRTDataset rasterXSize="2" rasterYSize="2">
  <GeoTransform>0.0, 2.0, 0.0, 0.0, 0.0, -2.0</GeoTransform>
  <VRTRasterBand dataType="UInt16" band="1">
    <SimpleSource>
      <SourceFilename relativeToVRT="0">{src_path}</SourceFilename>
      <SourceBand>1</SourceBand>
      <SrcRect xOff="0" yOff="0" xSize="4" ySize="4"/>
      <DstRect xOff="0" yOff="0" xSize="2" ySize="2"/>
    </SimpleSource>
  </VRTRasterBand>
</VRTDataset>''')

    read_vrt(vrt_path)
    # ValueError: could not broadcast input array from shape (4,4) into shape (2,2)

The upsample variant (SrcRect 2x2 -> DstRect 4x4) does not raise, but only the top-left 2x2 of each destination cell gets written; the remaining 12 cells stay at the fill value.

Root cause: xrspatial/geotiff/_vrt.py:464-472 computes a source-coordinate window from scale_x = sr.x_size / dr.x_size and scale_y = sr.y_size / dr.y_size, calls read_to_array(window=(src_r0, src_c0, src_r1, src_c1)), which returns an array sized to the source window, and copies it into the destination buffer at L577-583 with no resampling. The "size mismatch" guard at L573-575 compares src_arr.shape against out_r1 - out_r0 where out_r1 = out_r0 + src_arr.shape[0], so the guard never trips.

Fix plan: when sr.x_size != dr.x_size or sr.y_size != dr.y_size, resample the decoded src_arr from (sr.y_size, sr.x_size) to (dr.y_size, dr.x_size) before clipping and placing into result. Nearest-neighbour matches GDAL SimpleSource semantics. Carry the nodata mask through the resample. Clip after the resample so the existing window logic stays intact. No new dependencies; integer-ratio paths can use np.repeat or strided slicing, non-integer ratios use a per-axis nearest-index gather.

Regression tests will cover downsample 4x4 -> 2x2, upsample 2x2 -> 4x4, non-integer 3x3 -> 2x2, a per-band scale mix, window= on a downsampled source, and NoData preservation across the resample.