feat: make the backend dial timeout configurable (--backend-dial-timeout)

[wachirachoomsiri]

Follow-up to #569, which bounded the backend dial with a fixed 2s timeout so the auto-scale asleep MOTD is served promptly instead of hanging on the OS SYN-retry for a scaled-to-zero backend. This makes that bound configurable via a new --backend-dial-timeout flag (env BACKEND_DIAL_TIMEOUT), keeping the existing 2s as the default — so behavior is unchanged unless the flag is set.

Why

The ideal value depends on how quickly backends are reachable:

• On Kubernetes / in-cluster Services a healthy backend accepts the TCP connection in sub-millisecond time, so operators can lower the timeout (e.g. 100ms) to serve the asleep MOTD and trigger scale-up almost instantly on a scaled-to-zero Service.
• Deployments with slower or higher-latency backends can keep 2s (or raise it) to avoid treating a reachable-but-slow backend as asleep.

A single hard-coded constant can't serve both. #569 deliberately chose a conservative 2s; this leaves that as the default while letting operators tune it. (We run mc-router at 100ms on a Kubernetes Minecraft-hosting deployment where every backend is an on-cluster Service.)

Changes

• server/configs.go — add BackendDialTimeout time.Duration (default:"2s") to Config.
• server/connector.go — replace the backendDialTimeout constant with a per-Connector field (default defaultBackendDialTimeout = 2s) plus a UseBackendDialTimeout() setter, matching the existing UseAsleepMOTD / UseClientFilter style; the dial in findAndConnectBackend reads the field.
• server/server.go — wire config.BackendDialTimeout into the connector.
• README.md — document the new flag in the Usage list.

Testing

• go vet ./... and go test ./server/... ./mcproto/... pass.
• With no flag set, the dial uses the same 2s bound as before (defaultBackendDialTimeout).
• --help lists -backend-dial-timeout duration … (default 2s).

[itzg]

Released in https://github.com/itzg/mc-router/releases/tag/v1.44.1