fix(ntx-builder): keep alive main loop connection (#2362)

[SantiagoPittella]

Cherry pick of #2362 squash commit

Summary

Fixes the ntx-builder going permanently silent after a period of idle note traffic.

How:

1. Keepalive on every gRPC client.
2. bound stream.next() with a timeout so a stalled subscription is force-reconnected even if no error is ever surfaced.
3. Periodic log that escalates to a warning after no block is applied for a while, so a stall is a distinct signal instead of silence indistinguishable from an idle chain.

Changelog

[[entry]]
scope       = "ntx-builder"
impact      = "fixed"
description = "Network accounts no longer stop reactivating after idle periods: gRPC clients now use HTTP/2 keepalive to detect silently dropped block-subscription connections, plus a subscription stall watchdog and a liveness heartbeat."

[SantiagoPittella]

we need to avoid the squash merge in this PR I think cc @Mirko-von-Leipzig