Description

resolve_runtime_provider does not accept an api_mode parameter. When smart model routing routes a message to the cheap_model, the configured api_mode (e.g. anthropic_messages) is silently dropped, and the runtime defaults to chat_completions.

This breaks local inference servers that expect a specific API format, particularly when using Anthropic-format endpoints with SSD KV caching where the API mode determines cache hit behavior.

Steps to Reproduce

1. Configure smart routing with a cheap model that requires anthropic_messages API mode:

   smart_model_routing:
     enabled: true
     cheap_model:
       provider: custom
       model: local-model
       base_url: http://127.0.0.1:8000
       api_mode: anthropic_messages

2. Send a simple message that routes to the cheap model
3. Observe that the API call uses chat_completions format instead of anthropic_messages

Expected Behavior

api_mode from cheap_model config should take priority, falling back to the runtime-resolved value.

Actual Behavior

api_mode is always taken from runtime.get("api_mode"), ignoring the explicit config value.

Affected Component

Agent - Smart model routing (agent/smart_model_routing.py)

Platform

All platforms

Hermes Version

v0.8.0