fix: fix connection panic caused by WaitGroup misuse on close

[nodece]

Fixes #1483

Motivation

When the connection is closed while SendRequest/SendRequestNoWait are in flight, failLeftRequestsWhenClose calls Wait() on the WaitGroup. Concurrent goroutines that pass the getState() check may still call Add(1), triggering sync: WaitGroup is reused before previous Wait has returned panic in Go 1.25+.

Modifications

• Protect the state check and Add(1) under c.mu.RLock() in SendRequest and SendRequestNoWait. Close() sets state under c.mu.Lock(), so the check-and-add is atomic with respect to Close(). After state is set to connectionClosed, no new Add(1) can happen, ensuring Wait() will always return.
• Replace the nil-sentinel goroutine drain with a non-blocking drain after Wait(). Since Wait() guarantees all requests are already enqueued, a non-blocking drain is sufficient for strict correctness.