Updating config paths for MaxText on vLLM

[NicoGrande]

Description

Updates the config paths used in MaxText on vLLM codepaths and removes circular import in src/MaxText/__init__.py.

This PR also overwrites the KV cache variable in adapter.py to avoid duplicating KV cache instances in memory.

Tests

Running vllm_decode.py

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 0% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...t/integration/vllm/maxtext_vllm_adapter/adapter.py	0.00%	4 Missing ⚠️
src/MaxText/rl/train_rl.py	0.00%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[bvandermoon]

LGTM. Why was this causing duplicate KVCaches?

[NicoGrande]

LGTM. Why was this causing duplicate KVCaches?

This is probably taken care-of by Jax under the hood, but I wanted to avoid the possibility of having both the input KV cache and the updated KV cache in memory simultaneously in two separate variables. This is also how vLLM native models are implemented.