Remove workspace and flow labels from all infrastructure metrics
(consumer_state, processing_count, rate_limit_count, producer_count,
subscriber_state, received_count, dropped_count, tool_invocation_count)
to eliminate unbounded time-series growth from workspace × flow
combinations.

Wire ConsumerMetrics into ReceiverPool and ProducerMetrics into
SenderPool so that processing_count, request_latency, consumer_state,
and producer_count are emitted in the async pub/sub path — these
metrics were lost during the sync-to-async migration. Eagerly
initialise label sets on construction so metrics appear at scrape
time before the first message is processed.