feat: add OpenCode Zen as inference provider

[DheerGupta35959]

Summary

Adds OpenCode Zen as a first-class inference provider, enabling access to a curated selection of 75+ LLM models (GPT-5, Claude, Gemini, DeepSeek, Qwen, and more) through automaton's existing inference routing system.

Changes

• src/types.ts — Added "opencode" to ModelProvider union type, added opencodeBaseUrl/opencodeApiKey to config
• src/conway/inference.ts — Added OpenCode as a recognized InferenceBackend, routed through chatViaOpenAiCompatible with provider-specific error messages (401/429/404)
• src/inference/registry.ts — Added "opencode" to the providers that are skipped during baseline auto-disable (dynamically-discovered models manage their own lifecycle)
• src/index.ts — OpenCode provider startup: reads OPENCODE_BASE_URL/OPENCODE_API_KEY env vars, auto-discovers available models from proxy via GET /v1/models, populates the model registry
• src/config.ts — Added opencodeBaseUrl/opencodeApiKey to createConfig for setup wizard support
• src/__tests__/inference-router.test.ts — 4 unit tests covering model registry, auto-disable skip, model discovery, and fallback routing

Tests

Unit tests (68/68 passing):

• All existing tests plus 4 new OpenCode-specific tests pass.

Risk / Compatibility

• Additive only. No existing providers modified.
• OpenCode backend is opt-in via OPENCODE_BASE_URL and OPENCODE_API_KEY env vars.
• Auto-discovery from /v1/models populates available models at startup.

Example usage

1. Configure the OpenCode Zen connection:

export OPENCODE_BASE_URL=https://opencode.ai/zen/v1
export OPENCODE_API_KEY=sk-your-opencode-key

Or in your automaton config:

{
  "opencodeBaseUrl": "https://opencode.ai/zen/v1",
  "opencodeApiKey": "sk-your-opencode-key"
}

2. Any model served by your OpenCode Zen proxy is registered with provider: "opencode" and available for inference routing. The startup log shows:

OpenCode Zen: discovered 24 models from https://opencode.ai/zen/v1/v1/models

3. Model routing is automatic. Once registered, models route to the OpenCode backend based on the model registry lookup — no manual backend selection needed. The inference router picks the right backend (conway, openai, anthropic, ollama, or opencode) based on the model's registered provider.