Fix/metrics realtime

[meetj-videosdk]

Summary of Changes

models.py

• Set ttfb = e2e_latency = thinking_delay

realtime_metrics_collector.py

• Added mark_user_activity function for timestamping
• Added robust condition checking
• Added time parameter in timeline_event functions

realtime_pipeline.py

• Added on_last_audio_byte function to collect timestamps when the agent finishes its speaking

test_realtime_pipeline.py

• Added "text" modality in OpenAIRealtime for getting transcripts

realtime_api.py

• Added _user_speaking flag for tracking user speech
• Added safe_last_audio_callback function to remove unwanted interruptions
• Fixed timestamp collection for accurate TTFB calculation

live_api.py

• Added timestamp collection when the agent is interrupted

xai_realtime.py

• Added timestamp collection for TTFB calculation and when the agent is interrupted

aws_nova_sonic_api.py

• Added _user_speaking and _user_transcript_received flags
• Added timestamp collection for TTFB calculation and when the agent is interrupted