Proposal: add Runcrate as an AI provider

[heydryft]

Proposal: add Runcrate as an AI provider (language models, and later embeddings).

Runcrate (https://runcrate.ai) serves open-weight models such as DeepSeek V4.1 Flash, Qwen3, Kimi K2.6, GLM and gpt-oss behind an OpenAI-compatible API:

• Base URL: https://api.runcrate.ai/v1
• Auth: Authorization: Bearer <key>, env var RUNCRATE_API_KEY
• Model listing: GET https://api.runcrate.ai/v1/models (OpenAI format, includes modality info)

Following the "Add an AI Provider" playbook, step 0 is Esperanto support, so the plan would be:

1. A small PR to lfnovo/esperanto adding a runcrate language provider (OpenAI-compatible, same shape as the existing hosted ones).
2. Then in Open Notebook: a ProviderSpec with openai_compat_discovery_url, the SupportedProvider literal, PROVIDER_CONFIG, env_var_map, the discoverer, the test updates and the docs/CHANGELOG entries listed in the playbook.

Would you accept this? If yes, I'm happy to do both PRs. Disclosure: I'm using AI assistance to prepare the code, and I test it against the live API before opening anything.