geotiff tests: consolidate VRT tail cluster

[brendancol]

Closes #2437. Cluster 13 of the long-tail GeoTIFF test consolidation epic (#2424).

Summary

Fold the 12 VRT-tail residue files into existing vrt/ modules plus one new vrt/test_parity.py. No source changes; tests only.

Mapping:

• Missing-sources trio plus the public-default file -> vrt/test_missing_sources.py. The _1843 internal _vrt.read_vrt entry point and the XRSPATIAL_GEOTIFF_STRICT=1 env override are both preserved; _1860 adds the public read_vrt / open_geotiff('.vrt') default-raise coverage; _1799 and _2265 add the chunked-path policy (vrt_holes at build, raise-at-build, window / band scoping, multi-source preview).
• SrcRect-validation (_1784), kwarg-drop (_1685), tiled-validation (_1862) -> vrt/test_validation.py.
• Lazy-chunks (_1798) and direct read_geotiff_dask('.vrt') kwargs (_1795) -> vrt/test_window.py.
• Backend-parity (_2321), finalization-parity (_2162), backend-coverage (_2026_05_11) -> a new vrt/test_parity.py. These three are cohesive parity / coverage suites with their own harness, so they re-home as a group rather than fold into the matrix harness in parity/test_backend_matrix.py.

The _2321 parity helpers gained one .parent on the golden_corpus fixture path because the file moved from tests/ into tests/vrt/. Every other test is preserved with identical assertions; names dropped trailing issue numbers where the originating file namespaced them.

docs/source/reference/release_gate_geotiff.rst rows that cited the deleted files now cite the consolidated homes, so the checklist-parity gate in release_gates/test_stable_features.py stays green.

Audit

xrspatial/geotiff/tests/CLUSTER_AUDIT_VRT_TAIL.md maps every old file::test to its new file::test_id. It is deleted on a final commit on this branch before merge.

Verification

• pytest xrspatial/geotiff/tests/vrt/: 477 passed, 1 xfailed.
• pytest xrspatial/geotiff/tests/release_gates/test_stable_features.py -m release_gate: 158 passed, 1 xfailed.
• pytest xrspatial/geotiff/tests/: 5820 passed, 68 skipped, 2 xfailed.
• find xrspatial/geotiff/tests -name 'test_*.py' | wc -l: 175 (was 186; -12 deleted +1 added).

Test plan

• New vrt/test_parity.py passes.
• Consolidated vrt/ modules pass.
• Release-gate checklist-parity gate passes.
• Full geotiff suite passes.

[brendancol]

PR Review: geotiff tests: consolidate VRT tail cluster

Tests-only consolidation. I read the four touched test files in full, the new vrt/test_parity.py, the deleted originals on main, the release-gate checklist, and the checklist-parity gate that enforces it. Verified locally on the PR branch.

Blockers

None.

Suggestions

None blocking.

Nits

• vrt/test_missing_sources.py has test_build_raises_immediately in both TestChunkedMissingSourcesRaiseSmoke and TestRaiseAtBuild. The node IDs differ because they are class-qualified, so this is not a collision, and it mirrors the _1799 / _2265 split on main. Fine to leave; flagging only in case someone greps bare method names later.

What looks good

• Test count is conserved. vrt/test_parity.py collects exactly 55 nodes, matching the 55 across the three files it re-homes (_2321 + _2162 + _2026_05_11). The full geotiff suite passes with the same skip/xfail profile (5820 passed, 68 skipped, 2 xfailed), so nothing was silently dropped.
• The _GOLDEN fixture path in test_parity.py correctly gained one .parent for the move from tests/ to tests/vrt/; it resolves to the real golden_corpus/fixtures dir.
• The _1843 internal _vrt.read_vrt entry point and the XRSPATIAL_GEOTIFF_STRICT=1 env override are both preserved, as the issue asked.
• Release-gate .rst rows that cited the deleted files now cite the consolidated homes, and the checklist-parity gate (release_gates/test_stable_features.py -m release_gate) passes: 158 passed, 1 xfailed.
• The _1784 srcrect tests were ported from tempfile.TemporaryDirectory to tmp_path without changing behaviour; the relative-source VRT still resolves because the VRT and its source share the tmp dir.
• Helper functions are prefixed per fold (_raise_*, _srcrect_*, _kwarg_drop_*, _vrttail_*, _tiled_validation_*), so no cross-fold helper shadows another.
• Appended code is clean under the project's max-line-length=100.

Checklist

• No source changes (tests-only, as required)
• All implemented backends preserved (GPU tests stay skipif-gated)
• No tests silently dropped (node count conserved)
• Release-gate citations updated; checklist-parity gate green
• Audit file present; deleted before merge
• Temp / VRT filenames unique per test via tmp_path
• Full geotiff suite passes