fix(ui): initialize state selectively to prevent background task conflicts

[tusharmath]

Summary

Fix "JoinHandle polled after completion" panics caused by background tasks spawned during forge info racing with process exit.

Context

When running forge info, the handler was calling on_new() to initialize the model/agent state. However, on_new() also internally invokes hydrate_caches(), which spawns fire-and-forget background tasks. Because forge info exits quickly after printing output, these background tasks would still be running (or just completing) as the process terminated, triggering JoinHandle polled after completion panics in the async runtime.

The info command only needs agent/provider/model resolution — it has no use for the cache hydration side effects.

Changes

• Replaced on_new() call with init_state(false) in the TopLevelCommand::Info handler
• init_state(false) performs only the necessary state initialization without spawning background cache hydration tasks

Testing

# Run the info command and verify no panics occur
cargo run -- info

# Run with a specific conversation ID if applicable
cargo run -- info --conversation-id <id>

# Run the test suite for forge_main
cargo insta test --accept -p forge_main

Links

• Related: background task lifecycle management in UI::on_new() / UI::hydrate_caches()