Parallel query execution across backends

[renecannao]

Current distributed execution is sequential — queries to multiple backends are sent one at a time. Total latency = sum of all backend latencies.

Options:

• Thread per backend query (std::async / thread pool)
• Non-blocking MySQL/PgSQL APIs (mysql_real_query_nonblocking / PQsendQuery + PQgetResult)
• Event loop (epoll/kqueue)

The RemoteExecutor interface already abstracts this — swap the sequential implementation for a parallel one.

[renecannao]

Implemented: parallel open() for merge operators via std::async. Production callers pass parallel=true.