geotiff: fix flake8/isort style drift in tests and two source imports

[brendancol]

Closes #2970.

The geotiff subpackage has no flake8 or isort check in CI, so style drift had piled up. Running the project's configured tooling (setup.cfg: flake8 max-line-length=100, isort line_length=100) found a small set of issues, all confirmed by the tools. The 33 source files were already clean except for two import blocks.

What changed

• Cat 3 (pyflakes F-codes): removed dead test code.
  • tests/unit/test_geotags.py: dropped two redundant local import struct statements (F811) so the single module-level import is the one in use (clears F401).
  • tests/unit/test_metadata.py: removed the unused local ascii_buf (F841).
• Cat 1 (E501): tests/unit/test_metadata.py over-length import reflowed by isort.
• Cat 2 (W391): trailing blank line at the end of tests/unit/test_geotags.py.
• Cat 4 (isort): import ordering in __init__.py and _writers/eager.py, plus about 40 test files, normalized to the configured profile.

No behavioural change

Cat 1/2/4 are formatting only. The Cat 3 removals delete code that the tools confirm is never executed (dead imports and an unused local), so runtime is unchanged.

One detail worth flagging: three golden_corpus test files have imports below module-level setup that carry # noqa: E402. isort split those into one-per-line, which dropped the noqa from the second line, so I re-added # noqa: E402 to keep flake8 quiet on the intentional late import. isort is stable and idempotent after that.

Scope

No black/autopep8/ruff-format, no config widening. Fixes limited to what flake8 + isort report.

Test plan

• flake8 xrspatial/geotiff/ reports 0
• isort --check-only xrspatial/geotiff/ clean and idempotent
• pytest xrspatial/geotiff/tests/unit/test_geotags.py xrspatial/geotiff/tests/unit/test_metadata.py (171 passed)
• patched golden_corpus files collect and import cleanly

Found by /sweep-style.

[brendancol]

PR Review: geotiff flake8/isort style cleanup

Style-only change. No numerical, dask, or cupy code is touched, so the correctness/backend sections of the checklist are not applicable. I focused on the risk that actually matters for this kind of PR: did isort or the dead-code removal drop a symbol that is still needed?

Blockers

None.

Suggestions

None.

Nits

None.

What looks good

• The two source-file changes (__init__.py, _writers/eager.py) are pure import reflow to line_length=100. No symbol added or removed.
• The struct cleanup in test_geotags.py keeps the module-level import struct (still present at line 4) and drops only the two redundant function-local re-imports. struct is used throughout both functions via the module-level binding, so this is correct.
• The ascii_buf removal in test_metadata.py deletes a local that was assigned and never read.
• The # noqa: E402 handling on the three golden_corpus files is the one spot that needed care: isort split a parenthesized import into one-per-line, which would have stranded the second line without its noqa. Both lines now carry # noqa: E402, and flake8 is clean.

Verification

• flake8 xrspatial/geotiff/ reports 0; isort --check-only is clean and idempotent (no further changes on a repeat run).
• Full geotiff suite collects: 6208 tests, no import errors, so nothing was dropped by the reorder.
• pytest on the two directly edited unit files: 171 passed.

Checklist

• No behavioural change (formatting + confirmed-dead code only)
• No import symbol dropped (full-suite collection clean)
• flake8 and isort clean and idempotent
• Scope held to what the tools report (no formatter, no config widening)
• Algorithm / backend / NaN checks: not applicable (no compute code changed)