feat(llm): add Featherless AI provider

[aviflombaum]

Summary

Adds Featherless AI as a native LLM provider. Featherless hosts 17,000+ open-source models behind an OpenAI-compatible API, giving spacebot users access to models like Qwen3-32B, Llama, Mistral finetunes, and others not available through existing providers.

Registers featherless_key as a named key that auto-maps to a ProviderConfig with ApiType::OpenAiCompletions, following the same pattern as anthropic, openai, and openrouter. The generic provider dispatch in attempt_completion handles routing without any model.rs changes.

Changes

4 files changed, 55 insertions, 4 deletions

File	Change
src/config.rs	featherless_key field on LlmConfig, TomlLlmConfigFields, TomlLlmConfig; FEATHERLESS_API_KEY env var in load_from_env and from_toml; FEATHERLESS_PROVIDER_BASE_URL constant; auto-mapping to ProviderConfig in both config paths; has_any_key() check; needs_onboarding() env check; onboarding menu entry
src/llm/routing.rs	defaults_for_provider entry (default: featherless/Qwen/Qwen3-32B); provider_to_prefix entry
src/api/providers.rs	featherless field on ProviderStatus; all three endpoint handlers (get_providers, update_provider, delete_provider); routing auto-switch in has_key_for_current
src/api/models.rs	configured_providers discovery line

Usage

Config (config.toml):

[llm]
featherless_key = "your-api-key"

[defaults.routing]
channel = "featherless/Qwen/Qwen3-32B"

Environment variable:

FEATHERLESS_API_KEY=your-api-key

Model routing format: featherless/<org>/<model-name> — the featherless/ prefix is stripped before sending to the API, so model names match what Featherless lists (e.g. Qwen/Qwen3-32B, moonshotai/Kimi-K2.5).

Design notes

• No changes to llm/model.rs or llm/manager.rs — the generic ProviderConfig dispatch added in feat(llm): add custom providers and dynamic API routing #36 handles Featherless automatically once the key is registered with ApiType::OpenAiCompletions
• Uses or_insert_with so a custom [llm.provider.featherless] block in config.toml takes precedence over the named key auto-mapping
• Default model is Qwen/Qwen3-32B (32B parameter open-source model with tool calling support)

Why Featherless

Featherless specializes in hosting open-source models that aren't available (or are expensive) on other providers. Users get access to thousands of community and frontier open-source models with a single API key, complementing the closed-model providers already supported.

Testing

• Rebased on current main (no merge conflicts)
• Every change mirrors the existing named-key-to-ProviderConfig pattern
• No changes to existing provider behavior; all additions are purely additive