fix: rewrite SFTP connection pool for production resilience

- Add keepalive goroutine (5 min ping) to prevent Hetzner idle timeout
- getClient() returns nil on exhausted retries instead of dead client
- putClient() health-checks before returning to pool, auto-replaces dead connections
- Background fillPool() restores capacity without blocking callers
- Partial pool init allowed (no longer fails if 1 of 4 connections fails)
- Reader/Writer hold client until file closed (poolReadCloser/poolFileWriter)
- Pool.Alive() exposes current healthy connection count
- Pool.Close() for graceful shutdown

Author: troke12