feat: per-inbox model config and ollama lifecycle

[spignotti]

Summary

Allow each inbox to use a different LLM model (e.g. local Ollama for private folders, cloud model for fast processing). Optionally auto-start and stop Ollama when a local model is configured.

What Changed

• Per-inbox model override: InboxConfig now accepts optional model and api_base fields that override global [llm] settings
• OllamaConfig: New configuration section with auto_start, stop_after_run, host, and startup_timeout settings
• New ollama.py module: Lifecycle management with is_running(), start(), stop(), and ollama_lifecycle context manager
• extract_metadata(): Updated to accept explicit model/api_base params instead of LLMConfig object
• renamer.run(): Detects Ollama model usage and wraps execution in lifecycle context manager

Why

Different inboxes may require different models — private documents might use a local Ollama instance while general documents use cloud models. The lifecycle manager ensures Ollama is running when needed and stops it afterward.

Validation

• nox -s lint typecheck test — all pass
• 59 tests including 15 new Ollama lifecycle tests and 6 new model resolution tests
• Typecheck: 0 errors, 0 warnings

Config Example

[[inbox]]
path = "/path/to/private"
model = "ollama/llama3.2"
api_base = "http://localhost:11434"

[ollama]
auto_start = true
stop_after_run = true