feat: configure the connection pool and cap requests in flight (#146)

[tharropoulos]

• depends on fix: mark the node that answered as healthy (#144) #148
• refs AsyncClient locks up for good after a node restart under load: every call fails with PoolTimeout while the cluster sits idle #146, doesn't close it on its own: the lockup comes from a pool leak in httpcore (Cancelling requests under pool contention permanently leaks connection slots (AsyncConnectionPool → PoolTimeout) encode/httpcore#1093) that is unfixed upstream, so these settings make it less likely but can't remove it. a follow-up pr letting users pass an httpx2 client, which fixes the leak, will close it