(style) ruff pass: import sorting, unused imports, typing-stub cleanup

[rhoadesScholar]

Split out of #70 so the functional changes there stay reviewable. No behavior changes — this is a ruff --fix pass over the Python sources.

Stacked on v2.0_patch: merge #70 first, or retarget this to v2.0 afterwards.

What's here

44 files, all Python/stubs (no Rust, no docs, no test logic):

• Import sorting (I001) — stdlib / third-party / first-party grouping across daisy-py/python/daisy/, tests/, benchmarks/, examples/.
• 3 duplicate-import dedups — benchmarks/bench_dependency_graph.py, benchmarks/bench_worker_scaling.py, tests/test_worker_restarts.py each imported the same module twice.
• 9 unused-import removals (F401) — pytest from tests/daisy_compat/test_clients_close.py, tests/daisy_compat/test_dead_workers.py, tests/test_tcp_client.py; threading from examples/mws.py and tests/test_tcp_client.py; subprocess/sys from tests/test_context_passing.py; tempfile from tests/test_done_marker.py; Block from tests/test_tcp_client.py.
• 2 now-redundant # noqa removals (RUF100) — the suppressed rules no longer fire on those lines.
• Callable moved typing → collections.abc (UP035) in _daisy.pyi.
• Stub cleanups in _daisy.pyi — three redundant __repr__ declarations dropped (PYI029: stubs inherit object.__repr__), and timeout: float | int | object | None → float | object | None (PYI041: int is redundant beside float under the numeric tower).

Verification that it's semantically neutral

An AST-level comparison of every changed file's import set (name + alias, order-insensitive) before vs. after confirms the only net changes are the 9 removals, the 3 dedups, and the Callable module move — everything else is pure reordering. Each removed name was checked to have zero remaining references (the one lingering Block match in tests/test_tcp_client.py is inside a comment).

Full suites pass on this branch and on v2.0_patch identically: pytest tests/ 204 passed / 1 xfailed, cargo test -p daisy-core 46 passed.

Not in scope

The tree is not lint-clean after this — ruff check . still reports 74 findings, dominated by BLE001 (18 blind excepts) and S110 (15 try/except/pass), most of which are deliberate best-effort cleanup paths in the logging and worker-teardown code. There's also no ruff config in pyproject.toml and no lint CI job, so none of this is enforced going forward. Happy to follow up with a pinned [tool.ruff] section plus a CI check if you want the invariant to hold.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KxYd8bCjgZLoESSe3smbwF

[rhoadesScholar]

Rebased onto v2.0 and retargeted, now that #70 is closed in favour of #72 / #73 / #74. This is no
longer stacked on anything.

Two corrections to my original description, both found while doing this:

1. "or retarget this to v2.0 afterwards" was not actually possible as written. Two of the 44 files
existed only via #70 — build_wrapper.py and tests/test_worker_serialization.py — so a plain retarget
would have left hunks against files that aren't on v2.0. Regenerated the pass against v2.0 instead of
cherry-picking, so those two are simply absent: 42 files, +56/-99 (was 44 files, +59/-102).

2. The "74 findings" figure in my "Not in scope" section is not reproducible, and I should not have
quoted a bare number. There is no [tool.ruff] section and no pinned ruff version in this repo, so the
count depends entirely on the reviewer's local ruff default — an independent run with ruff 0.16.0
--isolated over comparable paths reports 143. Neither number is wrong; there is simply no shared
definition of "lint-clean" here yet.

This pass is now explicit about its rule set rather than relying on a default:

ruff@0.16.0 check --isolated --select I001,F401,F811,RUF100,UP035,PYI029,PYI041 --fix \
  daisy-py/python tests benchmarks examples
Found 69 errors (66 fixed, 3 remaining).

The 3 remaining need --unsafe-fixes (an F811 re-export in an __init__), which would change the
public API surface — out of scope for a lint pass.

Behaviour-neutrality, checked rather than asserted

Same environment, same deselect, with and without this commit:

with    3 failed, 194 passed, 1 deselected, 1 xfailed
without 3 failed, 194 passed, 1 deselected, 1 xfailed

Identical, so the pass changes nothing.

I also checked each removal that could plausibly have been load-bearing, since an unused-import removal
is only safe if the import had no side effect:

• tests/test_context_passing.py — pytest is still imported; only subprocess and sys go. (In
  the diff it looks removed because it moved.)
• tests/test_tcp_client.py — Block really is removed, and the only remaining Block match in that
  file is inside a comment (# Block 0 should have been retried).
• examples/mws.py — daisy.v2 and daisy.logging are still imported, just reordered; only
  threading goes. The file's argparse block runs before its main import cell, and there is no
  matplotlib.use() or equivalent, so the reordering is inert.

On those 3 pre-existing failures

They are on v2.0, not from this PR, and worth knowing since they will show up in anyone's local run:
daisy_compat/test_client.py::test_basic_migrated,
test_block_compat.py::test_status_mutation_propagates_from_compat_block, and
test_context_passing.py::test_shim_workers_get_distinct_identities — all three are fixed by #73
(they are the editable-install dill failure). Plus
test_run_stats.py::test_slowing_workload_reports_positive_slope, deselected above and fixed by #74.

So #73 + #74 together take the suite green; this PR is neutral on all of it.

The standing offer from the original description still holds: happy to follow up with a pinned
[tool.ruff] section plus a lint CI job, which is what would make this invariant actually hold going
forward.

[pattonw]

I think we both fixed this independently. I also added github action CI tools to cover this now so this should be much better going forward 🙏