[Claude] Bug haul: 7 bugs in `grid` and `core`

[cmdupuis3]

Version

v2026.07.0

How did you install UXarray?

Source

What happened?

Seven incidental bugs were found in scraping for performance issues. These line numbers are from the cmd/merge-OFE branch #1560, which will presumably be merged in soon.

Done	Site	Bug
	grid/geometry.py:895, :904	_check_intersection returns True (abort signal) where siblings return an intersection count. Numba unifies to int64, so True → 1, and the parity test at :838 counts the abort as one intersection — flipping the pole-inside-polygon verdict.
	grid/utils.py:432, grid/bounds.py:429, :431	np.isclose(a, b, tol) — the third positional parameter is rtol, not atol. Intended 1e-8 absolute becomes ~1.57e-8 effective on every pole test.
	core/dataarray.py:1066	weighted_mean on non-face/non-edge data warns, leaves weights=None, then calls weights.sum() at :1076 → AttributeError instead of the documented unweighted mean.
	core/api.py:587, :595, :599	concat compares grids by id(), so two structurally identical grids that went through Grid.copy() are rejected; and it returns the loop variable uxgrid rather than ref_uxgrid.
	core/utils.py:59, :61, :63	dim_ordered is assigned inside a loop and consumed after it — last-match-wins, and NameError if no variable matches.
	core/dataset.py:756	sel re-wraps with the unsliced self.uxgrid, so a sel on a grid dim leaves the grid inconsistent (isel does slice it).
	grid/connectivity.py:547	_face_nodes_to_sparse_matrix is CSR-sorted by construction and test-covered (test/test_helpers.py:190) but has no production callers — the CSR builder this design wants, already sitting unused. Its non_filled_element_flags return is a wasted float64 all-ones array.

What did you expect to happen?

Good things?

Can you provide a MCVE to repoduce the bug?

Multiple