[NNX] Checkpoint restore: error on weights missing from the checkpoint

[ecnal-cienet]

Description

Restoring a Linen-layout checkpoint into an NNX state filled every absent leaf with a default value. NNX-only rngs/dropout are legitimately absent from a Linen layout, but so is a weight that is genuinely missing from disk — and that weight was then silently zero-filled, a silent accuracy loss that also slips past the ShapeDtypeStruct guard (the default is a concrete zeros array, not a struct).

This distinguishes the two cases in _populate_pure_dict_from_partial by tracking the path and whether it sits under an rngs/dropout subtree:

• rngs/dropout absent → filled with a deterministic default, as before.
• any other absent weight → governed by the new checkpoint_missing_param_policy:
  • "error" (default): raise naming the parameter path and expected shape.
  • "warn": log a WARNING listing the zero-filled path and continue.

"error" is the default because a zero-filled weight is silent accuracy loss; opt into "warn" only when zero-filling is intended.

Tests

• tests/unit/checkpointing_nnx_load_test.py::TestMissingParamPolicy — present weights pass through; rng/dropout absence defaults even under error; a missing weight raises naming the path; warn zero-fills and logs; default policy is error.

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[codecov]

Codecov Report

❌ Patch coverage is 84.21053% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/maxtext/common/checkpointing.py	83.33%	3 Missing ⚠️

📢 Thoughts on this report? Let us know!