Fix mypy drift: cast xarray returns, drop deprecated numpy plugin

[thodson-usgs]

Summary

Recent mypy (1.20+) is stricter about no-any-return: functions declared
to return concrete xarray types can no longer silently return Any. Several
xarray operations (reindex_like, transpose, copy, rename, concat,
chunk bookkeeping) now come back typed as Any, tripping the check at 7
return sites across utils.py, methods/_shared.py, methods/interp.py,
and methods/conservative.py.

main last passed CI on 2026-02-06. If the lint job re-ran today against
current package versions it would fail on these 9 pre-existing errors, so
in-flight PRs (#70 / #71) can't go green without resolving the drift.

This PR also drops plugins = numpy.typing.mypy_plugin from [tool.mypy]
since that plugin is deprecated in numpy 2.3+ and slated for removal
(orthogonal to the no-any-return work, but cleaned up here for clarity).

Question to the maintainers — which fix do you prefer?

What this PR currently does is option A below, but it's not the only
reasonable choice and the commit is easy to rewrite. I'd appreciate a
preference before this is merged.

A. Add cast() at each of the 7 return sites (what this PR does).
Preserves warn_return_any = true — future leaks of Any back through
function boundaries still trip mypy, so the strictness is kept where
it's catching real bugs, with explicit acknowledgement at the known
false-positive sites.

• Pro: keeps the strictness knob on.
• Con: 7 cast(...) calls, a cast import in 4 files.

B. Drop warn_return_any = true globally from [tool.mypy].
One-line change, all 7 errors disappear. Consistent with the per-module
override that #70 already adds for conservative_2d for exactly this
reason — xarray's stubs don't narrow well enough for this check to be
net-positive project-wide.

• Pro: minimal code churn.
• Con: loses a legitimate check elsewhere.

C. Per-return # type: ignore[no-any-return] comments at each site.
Halfway between A and B — no cast import, but an inline waiver at
each site rather than a blanket silencing.

Happy to rewrite to B or C if either is preferred; just drop a comment.

Verified locally (option A)

• hatch run lint → green (ruff + mypy + ruff format --check)
• hatch run test → 71 passed, 1 skipped, 1 xfailed
• hatch build → sdist + wheel built

Test plan

• CI linting job passes
• CI build & test matrix passes on all OS/Python combinations

🤖 Generated with Claude Code

[thodson-usgs]

@BSchilperoort , I've left this PR in draft, but it notes three options for getting the CI passing. Do you have a preference? I don't have an opinion on this one.