Fix standalone DeepSeek L1 extraction

[sirenexcelsior]

Summary

Fix standalone L1 extraction compatibility with DeepSeek-style OpenAI APIs.

This PR makes two focused changes:

1. enableTools=false no longer passes read-only tools into generateText()
2. standalone gateway config can now opt into disableThinking, which injects
   thinking: { type: "disabled" } into the request body

Why

L1 extraction expects pure JSON text output. In practice, DeepSeek-compatible
backends can fail this path when:

• the runner still exposes tools for text-only extraction
• reasoning/thinking mode is enabled and the backend returns reasoning-related
  output that the current L1 parser cannot consume

Observed failures included:

No JSON array found in extraction response
The `reasoning_content` in the thinking mode must be passed back to the API.

Scope

• src/adapters/standalone/llm-runner.ts
• src/gateway/config.ts

Notes

This is intentionally small and scoped to standalone/gateway mode.
It does not change the OpenClaw-hosted path.

[Maxwell-Code07]

Thanks for this fix! We've noticed this issue as well. On certain reasoning models, the L1 extraction path can run into problems like what's described in #58. We'll review this and follow up with you soon.