Fix goroutine leak in reader on disconnect

[rradu92]

Problem

The inner goroutine in (*amiAdapter).reader blocks indefinitely on chanEvents <- event (or chanErr <- err) when the outer reader exits via <-stop. Nobody drains those channels after stop, and the inner goroutine has no way to observe stop directly, so it parks forever holding the bufio reader and the underlying socket.

Each AMI reconnect therefore leaks one goroutine.

Reproduction

A long-running service with Keepalive: true and intermittent network issues accumulated 100+ leaked reader goroutines visible in /debug/pprof/goroutine, all parked on chanEvents <- event at ami.go:429. In one production capture, 167 of 345 total goroutines (~48%) were stuck in this single stack — one per historical reconnect since process start.

167 @ ...
#       github.com/ivahaev/amigo.(*amiAdapter).reader.func1+0x1ac      ami.go:429

Fix

• Buffer chanErr and chanEvents with size 1.
• Use select-on-stop in the inner goroutine when sending, so it can bail out instead of blocking when the outer reader has already exited.
• In a defer, close the connection so any in-flight readMessage returns an error, then wait on innerDone to confirm the inner goroutine has fully exited before reader returns. This makes the cleanup deterministic for callers.

The outer reader's behavior is unchanged for the happy path; only the disconnect/stop path is now leak-free.

Tests

Existing tests pass (go test ./...).