Reduce Prometheus metrics overhead in batching scenario

[FZambia]

This PR improves CPU utilization under load in scenario when WriteManyFn is actively used (message batching scenario).

For bench run with throughput limited by publish rate (to make conditions equal):

go build && WRITE_DELAY=200 MAX_FRAME_MESSAGES=-1 ./bench_server
go run main.go -s ws://localhost:8000/connection/websocket -n 100000 -ns 500 -np 100 -pl 10 -p channel

This effectively reduced CPU usage from 56-60% -> 48-50% on my machine.

Before:

After (no incTransportMessagesSent at all):