feat(experimental): small model mode for local inference

[ericksoa]

Summary

• Adds small model mode for local inference providers (Ollama, vLLM) that reduces system prompt overhead so small local models have more context capacity for conversation
• Explicit ollama-local/vllm-local cases in getSandboxInferenceConfig() (was relying on default fallthrough)
• New NEMOCLAW_SMALL_MODEL_MODE build arg lowers bootstrap token budgets (bootstrapMaxChars=4000, bootstrapTotalMaxChars=8000) and writes compact workspace files (SOUL.md, AGENTS.md) at image build time
• Logs [experimental] during onboarding when active

Context

NVBUG 6018719: Ollama with qwen2.5:0.5b produces usable inference but garbage answers because OpenClaw's ~14KB+ default system prompt overwhelms the model's capacity. A/B testing showed the compact prompt saves ~1700 prompt tokens per turn — the difference between ~18 and ~30+ conversation turns before context exhaustion.

This is experimental — needs design review before graduating from draft.

Test plan

• Unit tests pass (479/479, 1 pre-existing timeout on main)
• A/B tested qwen2.5:0.5b locally: compact prompt produces equivalent quality answers with 97% fewer prompt tokens
• End-to-end: build sandbox with Ollama + small model, verify workspace files are written
• End-to-end: verify cloud providers (NVIDIA, OpenAI, Anthropic) are NOT affected
• Design review on compact workspace file content

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: f9e045b4-bc9c-42f9-ab26-c674838b2330

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/small-model-mode

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}