Refactor GeoTIFF Phase 5f: extract _encode.py from _writer.py (PR-L of #2211)

[brendancol]

Summary

Follow-up under epic #2211. Phase 5f deliverable: extract the write-side encode helpers (strip/tile encode + photometric + predictor + compression block) out of _writer.py into a new _encode.py. Mirrors the _decode.py extraction (PR-G) on the read side.

Scope

Mechanical extraction. Behavior-neutral.

Hold for dispatch until PR-K (_overview.py) merges — both PRs touch _writer.py and serial landing avoids a heavy rebase.

Files

• Create: xrspatial/geotiff/_encode.py.
• Modify: xrspatial/geotiff/_writer.py — remove the helpers being moved; import them from _encode.
• Modify: any module that imports those helpers (likely _writers/eager.py, _writers/gpu.py, _writers/vrt.py).

What to move

Encode helpers in _writer.py (verify each name in the current file post-PR-K):

• _invert_nodata_for_miniswhite
• _apply_photometric_miniswhite_invert
• _resolve_photometric
• _reject_disagreeing_photometric_override
• normalize_predictor
• _apply_predictor_encode
• _compression_tag
• _prepare_strip
• _write_stripped
• _prepare_tile
• _write_tiled
• _compress_block

Leave in _writer.py: _write, _write_streaming, _validate_lowlevel_write_kwargs, fsspec write helpers.

Acceptance

• _writer.py line count drops by ~500 lines. Combined with PR-I and PR-K, target ending around ~1300 lines.
• All encode-related tests pass unchanged: photometric, predictor, strip/tile encode, miniswhite invert, compression tag, plus the broader geotiff write suite.
• Public import paths preserved.
• No public API change.

Non-goals

• Splitting _write / _write_streaming further.
• Performance work.

Closes part of #2211.