Fix ESMF encode truncating the fill value into a valid node index

[cmdupuis3]

Closes #1745

Overview

_encode_esmf applies the UGRID→ESMF index offset to the whole face_node_connectivity array, padding included, and then encodes elementConn as int32. INT_FILL_VALUE + 1 is -2**63 + 1, and narrowing that to int32 truncates it to 1, so every padded slot of a ragged mesh is written to disk as node 0.

Nothing raises, and the resulting file is structurally valid, so any reader — including uxarray's own — takes those slots for real vertices. Faces gain spurious extra nodes and produce incorrect polygons. This is the case that originally surfaced this family of bugs.

The fix offsets only the valid indices and writes ESMF's own -1 into the padding, which is what the variable's _FillValue attribute already advertises. The test asserts on the encoded array and on what lands on disk after the int32 narrowing, rather than on a round trip: a round trip can hide an encode bug when the reader's inverse offset happens to cancel it.

PR Checklist

General

• An issue is created and linked
• Added appropriate labels (if your uxarray repo permissions allow it)
• Filled out Overview and Expected Usage (if applicable) sections

Testing & Benchmarking

• There is adequate test coverage of changes from this PR (add new tests if needed)

Documentation and Examples

• Docstrings updated with any function changes, and included in all new functions

AI Disclosure

AI Usage: Claude Opus 5

• I have tested and take responsibility for all AI-generated content in my PR.