Ambient mode: add provider_profile config option to route ambient to a different provider

Problem

Ambient mode always uses the default provider (e.g. Claude) regardless of what ambient.model is set to. There is no ambient.provider or ambient.provider_profile config option, so users cannot route ambient cycles to a different provider — such as a local model via [providers.<name>] — without changing the global default_provider, which would also affect interactive sessions.

Use case

I have a local Gemma 4 26B model running via llama.cpp with a [providers.gemma4] profile configured. I want ambient to use this free local model instead of burning Claude quota on cycles, while continuing to use Claude for interactive sessions.

Current config

[providers.gemma4]
base_url = "http://localhost:8080/v1"
api_key = "not-needed"
model = "gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf"

[ambient]
enabled = true
model = "gemma-4-26B-A4B-it-UD-Q4_K_XL.gguf"  # has no effect — still routes to Claude

Expected behavior

Setting ambient.model to a provider-profile model, or adding a new ambient.provider_profile field, should route ambient cycles through that provider instead of the default.

Proposed solution

Add one of:

• ambient.provider_profile = "gemma4" — references a [providers.<name>] block
• ambient.provider = "openai-compatible" + honor the model name to resolve the right profile
• Automatically infer the provider profile when ambient.model matches a model defined in a [providers.*] block

The first option (provider_profile) seems cleanest and most explicit.

Related

• #104 (ambient tool discovery bug — cycles are all Incomplete regardless of provider)

No files or tests are named; start by tracing ambient configuration parsing and the provider-selection path. Done means an explicit provider profile routes ambient cycles without changing the interactive default, while preserving the existing default-provider behavior.