Make vLLM instruct thinking mode configurable

[ArtemVazh]

Summary

• preserve the original change from Disable thinking mode in vLLM instruct chat #457 with its commit author intact
• expose enable_thinking on WhiteboxModelvLLM
• keep thinking disabled by default and forward the configured value through chat_template_kwargs

Attribution

The first commit is cherry-picked from #457 and retains dmndxld <dmndxld@gmail.com> as its author. The configurability change is a separate follow-up commit.

Supersedes #457.

Validation

• Python compilation passed
• isolated smoke checks covered both enable_thinking=False and enable_thinking=True
• no test files are included in the commits