fix: health dashboard poisoning from cancelled connectors

- FSW: report telemetry to health API (client_type=cloud-fsw)
- FSW: increase FANOUT_TIMEOUT 90s -> 180s, deploy timeout 120s -> 240s
- Connector-worker: return error_type/category/message in response JSON
- SDK engine: increase GLOBAL_TIMEOUT 90s -> 180s
- SDK engine: classify CancelledError as 'cancelled' not 'crash'
- SDK engine: skip cancelled connectors from telemetry (never ran)
- system_info: bump browser tiers (8GB->8, 16GB->12, 32GB+ capped at 12)