[codex] Fix GLM OCR token forwarding

[hansent]

Summary

This fixes truncated GLM OCR responses by addressing the token-limit path in both direct GLM inference and workflow remote execution.

What changed

• Restored the GLM-OCR model wrapper's default max_new_tokens behavior when requests arrive with max_new_tokens=None.
• Added max_new_tokens to the GLM OCR workflow block and forwarded it in both local and remote execution modes.
• Extended the SDK LMM client to serialize max_new_tokens and enable_thinking so remote workflow execution can pass generation parameters through to /infer/lmm.
• Fixed the Qwen3.5-VL workflow block's remote path to forward its existing generation parameters as well.
• Added targeted unit coverage for GLM default token handling, workflow remote forwarding, and SDK request serialization.

Root cause

There were two gaps working together:

1. LMMInferenceRequest can legitimately carry max_new_tokens=None, but the GLM-OCR wrapper forwarded that None directly into generation instead of restoring the model default.
2. The workflow remote execution path for GLM OCR did not expose or forward max_new_tokens, and the shared LMM SDK helper did not serialize that field either.

That combination made serverless workflow execution especially likely to use an unintended shorter generation limit.

Impact

• GLM OCR requests should no longer get cut short just because the token limit was omitted.
• Workflow remote execution can now intentionally override the token cap for GLM OCR.
• Qwen3.5-VL remote workflow execution now correctly forwards its generation settings too.

Validation

• python -m py_compile inference_models/inference_models/models/glm_ocr/glm_ocr_hf.py inference_sdk/http/client.py inference/core/workflows/core_steps/models/foundation/glm_ocr/v1.py inference/core/workflows/core_steps/models/foundation/qwen3_5vl/v1.py tests/workflows/unit_tests/core_steps/models/foundation/test_vlm_remote_execution.py tests/inference_sdk/unit_tests/http/test_client.py inference_models/tests/unit_tests/models/test_glm_ocr_hf.py
• git diff --check

Notes

pytest could not be run in this environment because the current Python environment does not have pytest installed (python -m pytest reported No module named pytest).