Close idle transport connections when targets are drained or disposed

[AnyCPU]

Each Target creates its own http.Transport with MaxIdleConnsPerHost: 100. When a target is replaced during deploy or removed via RemoveService, its idle connections are never explicitly closed. They linger until Go's default idle timeout (~90s).

During rapid deploys, these accumulate unnecessarily.

There are two disposal paths, neither of which cleaned up transport connections:

• Deploy (router.go:124-127, 153-156): Calls Dispose() then DrainAll() — in-flight requests are drained, but the transport's idle connection pool is never closed;
• RemoveService (router.go:193): Calls only Dispose() — no drain, no transport cleanup at all.

Fix

• Store the *http.Transport on the Target struct instead of creating it inline in the ReverseProxy;
• Call transport.CloseIdleConnections() at the end of Target.Drain(), after all in-flight requests have completed or been cancelled;
• Add TargetList.CloseIdleConnections() and call it from LoadBalancer.Dispose(), covering the RemoveService path where DrainAll is never called;
• Both calls are safe and idempotent — CloseIdleConnections only affects idle connections, not active ones mid-RoundTrip.