rpc: fix rare deadlock when canceling HTTP call context (#19715)

When cancelling the context for a call on a HTTP-based client while the
call is running, the select in requestOp.wait may hit the <-context.Done()
case instead of the <-op.resp case. This doesn't happen often -- our
cancel test hasn't caught this even though it ran thousands of times
on CI since the RPC client was added.

Fixes #19714

rpc/client.go

@@ -137,10 +137,12 @@ func (op *requestOp) wait(ctx context.Context, c *Client) (*jsonrpcMessage, erro
 	select {
 	case <-ctx.Done():
 		// Send the timeout to dispatch so it can remove the request IDs.
+		if !c.isHTTP {
 			select {
 			case c.reqTimeout <- op:
 			case <-c.closing:
 			}
+		}
 		return nil, ctx.Err()
 	case resp := <-op.resp:
 		return resp, op.err