[heydryft]

Adds Runcrate as a provider preset in AI settings.
Runcrate serves open models (DeepSeek, GLM, Qwen, Kimi, gpt-oss) behind an OpenAI-compatible API.
It uses the same pattern as #196 and #157.
It uses the existing openai path.
There is no backend change.

Changes

• src/views/setting/provider-form-dialog.vue: runcrate preset (apiCompatibility: 'openai', base URL https://api.runcrate.ai/v1), after Cheaper Inference.
• CHANGELOG.md: one line under Unreleased / Added.

How to test

1. Open Settings > AI > Add provider.
2. Select Runcrate.
3. Paste an API key from https://www.runcrate.ai/dashboard/api-keys?utm_source=sqlkit&utm_medium=directory&utm_campaign=gpu_pricing (keys look like rc_live_...).
4. Sync models.
5. Chat with deepseek-ai/DeepSeek-V4.1-Flash (model IDs carry the vendor prefix and are case sensitive).

Checks

• npx eslint src/views/setting/provider-form-dialog.vue: pass.
• npx vue-tsc --noEmit: pass.
• npx jest --ci --runInBand: pass (61 suites, 855 tests).
• Live, against the real API through data-studio-agent v0.2.0 (the version in src-tauri/Cargo.toml, the same harness calls agent_adapters.rs makes) with provider: "openai" and base URL https://api.runcrate.ai/v1: validate_llm_config returns true, list_llm_models returns 196 models, and run_agent_step returns a reply with finishReason: stop on deepseek-ai/DeepSeek-V4.1-Flash and on zai-org/GLM-5.3-Flash.
• Not run: the packaged Tauri app.

This PR was prepared with AI assistance: an AI coding agent wrote the change and ran the checks above.

Disclosure: I work on Runcrate.