[FEATURE REQUEST] Allow model selection for built-in operations when using local LLMs

Problem

When using Claude Code with local LLMs (Ollama, LM Studio, etc.), built-in operations fail because they try to use hardcoded Anthropic models:

• Session title generation: Always tries to use claude-haiku-4-5-20251001
• /insights command: Always tries to use claude-opus-4-5-20251101
• Other built-in agents: Hardcoded to use specific models

This causes request failures in local LLM environments.

Proposed Solution

Allow users to specify which models to use for different operations, similar to how --model works for the main session. This could be via CLI flags, config file, or model: inherit option.

Additionally, provide a fallback model option: Allow users to specify a fallback model that will be used when the specified model is unavailable, preventing request failures.

Related Issues

• #5146 - Feature Request: Allow model selection for built-in agents
• ollama/ollama#13939 - Timeout Error with session summary generation