Refactor `connect()` to Use Event-Driven D-Bus Signals Instead of Polling

[cachebag]

Overview:
The current connect() implementation uses polling with futures_timer::Delay to wait for device state changes and connection activation. This works but is inefficient and not idiomatic. NetworkManager emits D-Bus signals for state changes andwe should subscribe to these signals and react immediately rather than polling every 300ms.

Current Behavior:

• Polls device state in a loop with 300ms delays to detect disconnection
• Sleeps for 3 seconds after requesting a scan before proceeding
• Uses arbitrary timeouts and delays that may be too short or too long

Desired Behavior:

• Subscribe to NMDevice.StateChanged signal to detect disconnection immediately
• Subscribe to ActiveConnection.State and StateReason property changes to detect connection success/failure
• Use tokio::select! to wait on multiple signal streams simultaneously
• Provide detailed error messages when connections fail (wrong password, SSID not found, timeout, etc.)

References:

• NetworkManager D-Bus API - Device Interface
• NetworkManager D-Bus API - ActiveConnection Interface
• zbus Property Streams Documentation

Tasks:

• Add ActiveConnectionState and ConnectionStateReason enums to models.rs
• Implement wait_for_device_state() helper using receive_state_changed() signal
• Implement wait_for_connection_activation() helper using property change streams
• Replace polling loops in connect() with signal-based waiting
• Add proper timeout handling with tokio::time::timeout
• Test that connection failures provide meaningful error messages
• Remove all futures_timer::Delay usage from connect() method

This should theoretically give us faster response times, and definitely give us more robust error handling with specific failure reasons. This could also lower CPU usage (seemingly a given, since its just event-driven vs polling)

This follows NetworkManager best practices used by nmcli and other official clients. Definitely seems to be a non-starter if this should scale.