fix: adjust DeepSeek reasoning mode check condition

[JiajunBernoulli]

• Modified _drop_deepseek_incomplete_reasoning_history to properly handle reasoning mode detection
• Fixes issue DeepSeek-V4 reasoning_content error still reproducible on v0.1.5.post3 with WebUI + exec tool on Windows #3554

[Re-bin]

LGTM after the follow-up fixes.

The DeepSeek history cleanup is now scoped narrowly:

• Explicit thinking mode still cleans up incomplete reasoning history.
• Implicit cleanup only applies to known thinking-capable DeepSeek models such as deepseek-v4 / deepseek-reasoner.
• Non-thinking deepseek-chat keeps normal tool-call history intact.

Regression coverage was added for both the implicit deepseek-v4-pro case and the deepseek-chat non-trimming case.

Verified locally:

• uv run pytest tests/providers/test_litellm_kwargs.py -k 'deepseek' -q
• uv run ruff check nanobot/providers/openai_compat_provider.py
• git diff --check

Ready to merge once CI is green.