Skip self-copy when reading uncopied lazy-loaded field

[rayosborn]

Summary

When a file-backed NXfield is deep-copied (via NXdata.__init__, NXdata.weighted_data, etc.), NXfield.__deepcopy__ records _uncopied_data = (file, source_path) so the data can be lazily transferred to its new destination at next read. The transfer is done in NXfield._get_uncopied_data via f.copy(source_path, self.nxpath).

If the deep-copied field ends up at the same on-disk path as the source, h5py refuses the self-copy and raises RuntimeError: Unable to synchronously copy object (destination object already exists).

The case that surfaced this is the nexpy PlotDialog: when the user selects a field that already lives inside an NXdata (e.g. the summed_data_mask companion field), the dialog wraps it in a fresh in-memory NXdata named after the original group and reparents to the original grandparent. The deep-copied wrapper's nxpath collides with the source on disk, so the read crashes.

The fix is a one-line guard in _get_uncopied_data: when source and destination paths agree, the data is already where it needs to be — skip the copy and fall through to the normal readvalue. It addresses the symptom directly and also closes the door on any other code path that wraps a file-backed lazy field into a same-on-disk-path container, not just PlotDialog.

Test plan

• New tests/test_files.py::test_read_lazy_field_after_same_path_rewrap reproduces the original RuntimeError (uses a 2k × 2k mask so the field stays lazy-loaded — small fields skip the bug because their _value is already in memory and __deepcopy__ clears _uncopied_data). The test reads cleanly after the fix and asserts the _uncopied_data reference is cleared after the read.
• Full suite still passes (pytest tests/ — 104 tests).
• Confirmed in nexpy: plotting a mask field (or any field contained in an NXdata) via the PlotDialog no longer raises Unable to synchronously copy object.