fix(network): close peer connections that go idle (backport of #4562)

[stevenvegt]

Backport of #4562. Follows the backport of #4467 (#4564 on this branch), which fixed the other half of the same incident.

Problem

A peer kept a stream open for our node after our node had already dropped it, so it believed it was still connected and rejected every new connection with already connected. In our case the stale stream was held by an nginx in front of the peer; a half-open TCP connection does the same. Nothing cleaned that up short of restarting the proxy.

Change

New network.idletimeout (default 2m, 0 disables). Peers send gossip and diagnostics on every stream every few seconds, so a stream with no incoming message for two minutes is dead. The node now closes it and the regular reconnect logic takes over. Time spent handling a message (a large transaction list during sync, for example) does not count as idle. This works through reverse proxies, unlike gRPC keepalive pings, which proxies answer themselves.

The second part of the original PR is a refactor with no behaviour change: the three places that start a goroutine for a connection now go through startGoroutine or startReceiveLoop, so the goroutine counter and the receive-loop WaitGroup are maintained in one place instead of by hand at each site.

Conflicts

• docs/pages/release_notes.rst: this branch keeps its own Unreleased section. Resolved by adding the idle-timeout entry in this branch's style, below the three entries from fix(network): back off when a peer rejects the connection (backport of #4467) #4564.
• docs/pages/deployment/server_options_didnuts.rst does not exist on this branch. Dropped it; server_options.rst and cli-reference.rst were regenerated instead and carry the new option.

The code applied without conflicts, including the /v5 module paths, because the changed hunks do not touch import lines.

Verification

go test -race ./network/... passes, apart from TestNetwork_checkHealth, which fails locally for environment reasons on this branch without the change as well. Generated docs are up to date.

Note on TestNetworkIntegration_OutboundConnection11Reconnects: it failed twice during the full-package runs while preparing this backport. It is not caused by this change (6 clean full-package runs with it, 4 without it, 10 clean runs of the test in isolation). The test waits defaultTimeout (5s) for a reconnect while a clean disconnect sets a random 1 to 5 second backoff, so the budget is too tight under load. Pre-existing, same family as #4321.

Operational notes

For deployments behind nginx, grpc_read_timeout and grpc_send_timeout at their 60s default (not raised to hours) plus so_keepalive on the listen directive let the proxy drop dead streams on its own as well.