Migrate LlamaDecoderLayer to NNX

[RissyRan]

Description

This is co-work with @bvandermoon and @shuningjin

Example to migrate llama2 to NNX. This change is based on @bvandermoon's branch and keep it untouched for sanity check. This PR squashes all commits into one.

Tests

1. unit test can pass without logit mismatch
2. can run decode.py without shape mismatch. With checkpoint, it is giving reasonable output (log)
3. can run JetStream with PDB=18 on v6e-8, suggesting the memory issue might have been resolved. Here is the output comparison with main.

Checklist

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

• I have performed a self-review of my code.
• 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.