Team Performance page opens a WebSocket per instance, does not scale

The Team Performance page opens a separate WebSocket connection for every running instance in the team. For teams with 100+ instances this means 100+ concurrent WebSocket connections from a single browser tab, which puts unnecessary load on both the client and the server.

The same pattern exists to a lesser degree with instance status polling on listing pages, where each instance gets its own independent polling timer hitting the REST API individually rather than using a single batched request.

## Expected Behavior

A single connection (or a small fixed number of connections) should be sufficient to stream live resource data for all instances on the page, regardless of how many instances are listed.

## Suggested Direction

The frontend already has a well-structured MQTT service with connection pooling, subscription management, and reconnect handling — currently used for device logs and device resources. This service could be extended to also cover instance resource streaming by subscribing to the relevant topics through a single shared connection, rather than opening a raw WebSocket per instance.

This would bring instance performance monitoring in line with how device monitoring already works and avoid the N-connections-per-page scaling problem entirely.