fix: improve WS event delivery and add breakpoint node glow

[cristipufu]

Summary

• Serialize WebSocket sends per connection using an asyncio Queue + consumer task, preventing concurrent send_json calls on the same socket that could corrupt frames or silently fail
• Remove redundant run.updated broadcasts from handle_trace and handle_log — each trace/log was triggering both its own event AND a full run.updated, doubling message volume for fast scripts
• Conditional REST refetch on terminal status — when a run completes/fails, only refetch via REST if the store's trace/log counts don't match the run summary (recovers any missed WS events without causing canvas flicker)
• Isolate WS handler errors — wrap each handler in its own try-catch so one failure doesn't swallow the message for all handlers
• Red pulsing glow for breakpoint-hit nodes — isPausedHere nodes now use var(--error) border + node-pulse-red animation, matching the green pulse on executing nodes

Test plan

• Run a fast script (no I/O, completes in <100ms) and verify all traces/spans appear without needing a page refresh
• Run a slower script and verify real-time trace streaming still works
• Set a breakpoint and verify the paused node has a red pulsing glow
• Verify executing nodes still show green pulsing glow
• Verify no canvas flicker when a run completes

🤖 Generated with Claude Code