Configure vLLM instruct and thinking modes

[ArtemVazh]

Summary

• always initialize WhiteboxModelvLLM with instruct=True
• derive the adapter’s enable_thinking value from ThinkBooster’s existing model.disable_thinking_mode setting
• avoid introducing a separate instruct configuration option in this reasoning-focused repository

The thinking-mode mapping reuses the existing setting as the single source of truth:

• disable_thinking_mode: false → enable_thinking=True
• disable_thinking_mode: true → enable_thinking=False
• missing value → disable_thinking_mode=False → enable_thinking=True

Dependency

Depends on IINemo/lm-polygraph#464, which adds the configurable enable_thinking constructor argument. Merge that PR first.

Validation

• Python compilation passed
• git diff --check passed
• no test files are included in the commit