Add support for nullable API keys and LM Studio

[sohamb117]

Addresses #3185.

Adds support for nullable API keys and LM Studio as a first-class provider, improving the experience for local LLM users.

Key Changes

• Nullable API keys: apiKey can now be set to null for local providers that don't require authentication (previously required dummy strings like "no-key" or "dummy")
• LM Studio provider: Added lm_studio as a built-in provider with auto-detection via port 1234
• Documentation updates: Updated examples for Custom Provider, vLLM, and added comprehensive LM Studio
  setup guide

Motivation

Local LLM servers (LM Studio, vLLM, llama.cpp, etc.) don't require API keys, but the previous schema required a non-empty string. This forced users to use dummy values like "dummy" or "no-key", which was confusing and inelegant. Additionally, LM Studio is a popular desktop app for running local models, and adding it as a first-class provider improves discoverability.

Changes

Core Changes

nanobot/config/schema.py:

• Changed ProviderConfig.api_key type from str to str | None to support null values
• Added lm_studio field to ProvidersConfig

nanobot/providers/registry.py:

• Added lm_studio provider spec with:
  • Aliases: lm-studio, lmstudio, lm_studio
  • Default base URL: http://localhost:1234/v1
  • Auto-detection via port 1234
  • Backend: openai_compat

Documentation Updates

README.md:

• Custom Provider: Updated guidance to recommend apiKey: null instead of dummy strings for local servers
• vLLM: Updated example to use apiKey: null instead of "dummy"
• LM Studio: Added comprehensive setup guide with:
  • Installation instructions
  • Server startup steps
  • Configuration examples
  • Usage notes
• Provider table: Added LM Studio to the list of supported providers

Test Plan

• Verify apiKey: null works with existing local providers (ollama, vllm, custom)
• Verify LM Studio provider auto-detection with http://localhost:1234/v1 endpoint
• Verify explicit provider: "lm_studio" configuration works
• Verify backward compatibility: existing configs with non-null API keys still work
• Test with actual LM Studio instance (if available)
• Verify documentation examples are accurate

Breaking Changes

None. This is backward compatible:

• Existing configs with string API keys continue to work
• The change is purely additive (nullable type extends the existing string type)

[Re-bin]

LGTM ;)

[sohamb117]

Thanks!