fix(redis): ping the subscription connection to detect half-open sockets

[LounisBou]

The Redis transport opens two connections from the same DSN, one for commands and one for the pub/sub subscription. The periodic ping only reached the command connection, so a half-open subscription socket went unnoticed: the hub kept answering publish requests while every subscriber stopped receiving anything.

The ping now runs on both connections. Paired with a non-zero readTimeout it turns a silent socket into a rejected promise, which reaches Hub::die and lets the supervisor restart the hub.

This is what kept four production hubs frozen for 35 hours while the process stayed alive and the Redis instance stayed up.

Covers the ping method with two tests, so its coverage-ignore marker is gone.