Audio-gated flows and flow picker improvements

What was built & why

Static states stalled in text mode because they waited for an 'audio_done' event that never fired. The flow picker also crashed by loading invalid '_routing' config partials. You added _advance_text_autostates to auto-execute read-only tools and mimic the voice engine. You updated API endpoints in web_server.py to filter out underscore-prefixed files. This keeps text flows moving without breaking the state manager.

Principles

• Simulate missing events for non-interactive modes instead of blocking execution.
• Hide internal or partial configuration files from user-facing lists.
• Mirror live engine behavior in static modes to reduce divergence.

Patterns

• Auto-advance simulation for gated states

• Prefix-based filtering for internal assets

How to apply

• Apply _advance_text_autostates to any flow step that requires audio confirmation but runs in text-only contexts.

• Filter any UI list by excluding items starting with an underscore to prevent loading partial configs.

Pitfalls

• Auto-advancing may skip intended pauses if you apply it to interactive tools.
• The hardcoded max_hops limit of 16 can truncate long non-interactive sequences.
• Coupling text mode to static vs act state details makes refactoring harder.