Objo applications can interact with local and hosted LLMs

Add a standard-library API that lets Objo developers build conversational and agentic applications with large language models. It should support local or self-hosted models exposed through OpenAI-compatible endpoints, together with the hosted APIs from OpenAI, Anthropic, and Google Gemini.

The API should make simple chat requests straightforward while also supporting richer multi-turn workflows. Objo applications should be able to:

• Configure a provider, model, endpoint, and application-supplied credentials.
• Compose and retain system, user, assistant, and tool messages.
• Send requests and receive normal or streamed responses asynchronously.
• Declare tools, receive structured tool calls, execute them in application code, and return tool results to the model.
• Access provider-supported thinking or reasoning content, finish reasons, token usage, and other useful response metadata.
• Handle cancellation, timeouts, authentication failures, rate limits, malformed responses, and other provider errors consistently.
• Use provider-specific capabilities when they cannot be represented by the common API without weakening the provider-neutral experience.

The result should give hobbyist developers an approachable path from a minimal chat example to capable tool-using applications without requiring them to implement HTTP protocols or provider payload formats themselves.