Fix otlp session fragmentation

[krisztianfekete]

This PR fixes OTLP session fragmentation when some spans in a trace lack gen_ai.conversation.id (e.g. kagent's A2A infrastructure spans vs agent spans):

1. Pre-scan batch for conversation_id (otlp_routes.py): Before processing any spans in an OTLP batch, scan all spans across all scopes to find gen_ai.conversation.id. This ensures infrastructure spans (no conv_id) and agent spans (with conv_id) that arrive in the same HTTP request all route to the same session.
2. Route active sessions by trace_id + absorb orphans: When a span's trace_id is already in an active session (under a different name), route there instead of creating a new session. When routing to a conversation_id session, absorb any existing orphan session for the same trace_id (merging its spans/logs, cancelling its timers, removing it from the session map).
3. UI cleanup: Added session_removed event so the frontend removes absorbed orphan sessions from the session list.