reconnect() holds the pool slot while sleeping between attempts

`Session::reconnect()` sleeps `retry_interval` between passes and walks all endpoints while still holding its pool slot. With defaults (3 endpoints, 3 attempts, 1s interval) the worst case is ~92s, during which concurrent `acquire()` callers get "pool exhausted" — an error that points away from the real cause.

Fix: when the session is handed out from a pool, skip the between-attempt sleeps (the slot is still held, but the hold time drops to the connect+handshake attempts), or shorten them. Standalone sessions keep the current C#-style pacing.

Start by reading Session::reconnect and the pool path that hands sessions to acquire callers, focusing on how pooled versus standalone sessions are identified. Verify that pooled reconnect attempts do not retain the between-attempt sleeps while standalone sessions keep their current pacing, and confirm that concurrent pool acquisition no longer fails because of the delay.

Written by the indexing model from the issue text.