Add BigTIFF COG external compliance test and tier entry (#2303)

[brendancol]

Summary

• Adds xrspatial/geotiff/tests/test_bigtiff_cog_compliance_2286.py, the external-interop gate for the BigTIFF + COG writer combination. Per-row contract mirrors test_cog_writer_compliance.py (COG readiness: external compliance test suite for to_geotiff(cog=True) (#2286 PR 2/6) #2292): rasterio.open, byte-exact base pixels, overview shape and count, CRS/transform/nodata/dtype/band survival, COG layout invariant (IFDs before tile data), optional rio-cogeo validator with the existing skip pattern.
• Matrix is scoped to the BigTIFF-specific layout rather than re-validating every codec: one lossless integer codec (deflate / uint16), one lossless float codec (zstd / float32), single-band and 3-band, one overview level, plus a NaN-nodata row and the optional validator row.
• Auto-BigTIFF threshold row monkeypatches _compute_classic_ifd_overhead to drive estimated_file_size > UINT32_MAX from a 64x64 raster rather than allocating a true >4 GiB array. Mirrors the strategy in test_eager_bigtiff_overhead_exact_1905.py.
• Adds writer.bigtiff_cog: advanced to SUPPORTED_FEATURES at the end of the dict (Promote local COG contract to stable: flip tiers, document the contract, release notes (#2286 prod-ready wave A) #2300 is editing the writer.cog / reader.local_cog / reader.http_cog entries in parallel, so this lands clear of that diff).
• Adds a BigTIFF COG (issue #2303) section to docs/source/reference/geotiff.rst covering the threshold, the tier, and a pointer to the new test file.

Tier decision

All matrix rows pass locally. Per the wave D rule the tier stays advanced in this PR; promotion to stable happens after the gate has lived in CI for a release cycle. No rows triggered a real BigTIFF bug, so no xfail markers were needed.

Test plan

• pytest xrspatial/geotiff/tests/test_bigtiff_cog_compliance_2286.py -- 6 passed, 1 skipped (validator skip is expected when rio-cogeo is not installed)
• pytest xrspatial/geotiff/tests/test_cog_writer_compliance.py xrspatial/geotiff/tests/test_eager_bigtiff_overhead_exact_1905.py xrspatial/geotiff/tests/test_features.py -- 163 passed, 5 skipped
• from xrspatial.geotiff import SUPPORTED_FEATURES; SUPPORTED_FEATURES['writer.bigtiff_cog'] == 'advanced'
• CI green on the runner

Closes #2303.

[brendancol]

Self-review

Looked over the diff before opening for external review. Findings:

• Per-row assertions match the COG readiness: external compliance test suite for to_geotiff(cog=True) (#2286 PR 2/6) #2292 shape (rasterio.open + tiled + dtype tuple + byte-exact base pixels + overview factor + overview shape + CRS + transform + IFDs-before-data). Each row also asserts the on-disk magic is 43 and that parse_header reports is_bigtiff=True, so the BigTIFF-specific paths get covered rather than just inferring from bigtiff=True at the writer call.
• The auto-threshold row uses the same _compute_classic_ifd_overhead monkeypatch precedent that lives in test_eager_bigtiff_overhead_exact_1905.py. Re-allocating a multi-GB raster in CI was off the table, so this is the lightest path that still drives the production decision through the same helper that _assemble_tiff calls.
• SUPPORTED_FEATURES['writer.bigtiff_cog'] lands at the end of the dict with an inline comment that points at the test file and the docs section. Out of the way of the Promote local COG contract to stable: flip tiers, document the contract, release notes (#2286 prod-ready wave A) #2300 sibling edits.
• Docs paragraph sits below the VRT missing sources block, clearly separated from any future Stable COG contract section Promote local COG contract to stable: flip tiers, document the contract, release notes (#2286 prod-ready wave A) #2300 might add.

Did not promote the tier to stable in this PR (per the wave D rule). No row uncovered a real bug, so no xfail markers were needed.