Fix audio eMOS calculation and display

[Mupati]

Summary

• Fix v1_summary to extract inline audio inbound tracks from main events — inbound audio tracks were embedded inline on main stat events but never iterated over, so compute_emos was never called for them. Added build_track_data helper and loop over inline tracks per snapshot.
• Fix compute_emos packet loss fallback — previously required remote-outbound-rtp data (which many browsers don't send) for packetsSent. Now falls back to packetsLost / (packetsLost + packetsReceived) from local inbound-rtp stats when remote data is unavailable.
• Fix compute_emos RTT source — changed RTT lookup from remote.audio.outbound (which never carries roundTripTime) to remote.audio.inbound.
• Fix compute_emos one-way delay — changed delay = RTT to delay = RTT / 2 per E-model specification.
• Fix compute_emos guards — added bitrate 0 guard before math.log, packets_sent 0 and total_packets == 0 guards, and fixed operator precedence in auxiliary delay (0.1 * (delay - 150)).
• Fix check_emos_score shared mutable state — replaced shared issue_template dict with a make_issue_template() factory function to prevent cross-participant corruption.
• Fix check_emos_score running average — replaced incorrect (avg + score) / 2 with Welford's online mean.
• Fix event ordering — added ORDER BY created_at ASC, id ASC to get_stats_events and removed events.reverse().
• Add pending track buffer — per-track events that arrive before their connection's main event are queued and flushed when the snapshot is created.
• Add @staticmethod to get_stats_events.

Test plan

• Verified eMOS scores computed for Chrome and Edge participants across multiple test calls
• Verified packet loss fallback produces correct scores when remote-outbound-rtp is absent
• Verified zero errors in API logs during summary building
• Verified check_emos_score correctly accumulates per-participant averages