fix(parse): reject unsupported local models

[ChrisJar]

Summary

• reject unsupported local Nemotron Parse model IDs during ExtractParams validation
• keep explicitly configured remote Parse model contracts available
• document that NeMo Retriever 26.08 supports NVIDIA-Nemotron-Parse-v1.2 for local extraction

This prevents a Parse 2.0 request from being silently ignored while the local actor loads Parse v1.2.

Validation

• pre-commit run --all-files
• 27 focused parameter and local Parse tests passed
• git diff --check

[greptile-apps]

Greptile Summary

The PR prevents unsupported local Nemotron Parse model IDs from being silently accepted while preserving explicitly configured remote model contracts.

• Defines the supported local model as NVIDIA-Nemotron-Parse-v1.2.
• Validates local Parse model selection during ExtractParams construction.
• Adds coverage for rejected local models and accepted remote models.
• Documents the supported model and public Pydantic validation behavior.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
docs/docs/extraction/nemo-retriever-api-reference.md	Documents the supported local Parse model, remote-model configuration, and correct public ValidationError contract.
nemo_retriever/src/nemo_retriever/common/params/models.py	Rejects non-default Parse model IDs only when extraction is configured locally.
nemo_retriever/src/nemo_retriever/common/params/utils.py	Introduces the canonical supported local Nemotron Parse model identifier.
nemo_retriever/tests/test_params_models.py	Covers unsupported local model rejection and remote model acceptance through both endpoint fields.

_{Reviews (2): Last reviewed commit: "docs(parse): correct validation exceptio..." | Re-trigger Greptile}