geotiff: reject non-string compression at the to_geotiff boundary (#2975)

[brendancol]

Closes #2975

to_geotiff raised a raw AttributeError: 'int' object has no attribute 'lower' when compression was a non-string (say an int) and compression_level was set. The public wrapper validated compression names only inside an if isinstance(compression, str): block, so a non-string skipped validation and later hit compression.lower() in the compression_level range check.

• Add a type guard at the public boundary in eager.py that rejects a non-string, non-None compression with a TypeError, matching the low-level writer's existing guard in _writer.py.
• The string-name ValueError path and the None exemption are unchanged.

Backends: validation happens in the shared public wrapper before backend dispatch, so this covers numpy, cupy, dask+numpy, and dask+cupy.

Test plan:

• New regression tests assert to_geotiff(arr, path, compression=<non-str>, compression_level=1) and the no-level variant raise TypeError instead of AttributeError, parametrized over int/float/list/object.
• A positive test confirms the default, the 'none' string, and a string + compression_level combo still write.
• Full write/test_basic.py passes (289 tests); flake8 clean on changed files.

[brendancol]

PR Review: reject non-string compression at the to_geotiff boundary (#2975)

Blockers (must fix before merge)

None.

Suggestions (should fix, not blocking)

None.

Nits (optional improvements)

• eager.py:499 -- The guard exempts None (and compression is not None) to mirror the low-level writer's guard, which is the right call for keeping the two shapes aligned. One thing to note: to_geotiff(arr, path, compression=None) still leaks a downstream AttributeError from _compression_tag (_encode.py:311), because the signature types compression as str (default 'zstd') and None was never a supported value here. That is pre-existing and out of scope for this PR. I'm flagging it only so None isn't read as a value the new guard now accepts. A follow-up could either treat None as an alias for 'none' or reject it at the same boundary.

What looks good

• The guard sits before every compression.lower() call (the name check, the JPEG and experimental gates, and all three compression_level range checks), so no non-string can reach .lower().
• The TypeError type and message match the low-level writer's guard in _writer.py:215, so the public and private boundaries stay consistent.
• Tests cover the reported case plus int/float/list/object, with and without compression_level, and a positive test confirms the default, the 'none' string, and a string + level combo still write. Temp files carry the issue number.
• Validation runs in the shared wrapper before backend dispatch, so the fix applies to numpy, cupy, dask+numpy, and dask+cupy without per-backend changes.

Checklist

• Algorithm matches reference/paper: n/a (validation fix)
• All implemented backends produce consistent results: yes (pre-dispatch validation)
• NaN handling is correct: n/a
• Edge cases covered by tests: yes (int/float/list/object, level/no-level, valid codecs)
• Dask chunk boundaries handled correctly: n/a
• No premature materialization or unnecessary copies: yes
• Benchmark exists or is not needed: not needed
• README feature matrix updated: not needed (no new API)
• Docstrings present and accurate: unchanged; compression already documented as str