fix(pool): add connect timeout — prevent 20-second stall on Deluge reconnect

[wcypierre]

Problem

The torrent list was taking 20+ seconds to load. Root cause: the pool worker goroutine was blocking indefinitely inside conn.Connect().

pool.getConn() called conn.Connect() synchronously. The pool has a single worker goroutine — while it is inside getConn(), the entire select loop in worker() is frozen. All concurrent HTTP requests waiting for a pool connection are also stuck (they cannot send on pool.get until the pool worker is ready to receive).

conn.Connect() in go-libdeluge uses new(net.Dialer) with no dial timeout. When the TCP connection hangs (no RST, just silence), Go waits for the OS SYN timeout — ~20 seconds on Windows.

Why it regressed now

The April 26 commit changed go.mod from go 1.17 → go 1.24, then the telemetry commit raised it to go 1.25.0. From Go 1.21+, the go directive activates the pure Go DNS resolver, which returns both ::1 and 127.0.0.1 for localhost. If the Deluge daemon listens only on IPv4 and the system drops (rather than refuses) IPv6 packets, the SYN to ::1:PORT silently times out after ~20 s before Go falls back to IPv4.

With IdleConnectionTime = 30s, connections expire if idle for 30 seconds. Each reconnect event then triggers the 20-second hang, blocking all concurrent requests.

Fix

Run conn.Connect() in a goroutine and select on the result against a configurable timeout (default 10 s, overridable via --connect-timeout / POOL_CONNECT_TIMEOUT). Orphaned Connect() goroutines on timeout are cleaned up asynchronously.

The pool worker is still serialised (one connection established at a time), but it can no longer be stalled for longer than ConnectTimeout.