fix(bigquery/storage/managedwriter): faster context failure on send

This is a minor PR that accelerates the failure loop for context
expiration cases.  Previously, the connection abstraction would
deal with expired context on the receiver side, and this change
simply checks context before allowing a new append to succeed.

Tests already existed that exhibited this behavior, but were timing
dependent and masked the issue (the receiver would close the connection
fast enough).

Fixes: googleapis#10128

bigquery/storage/managedwriter/connection.go

@@ -343,6 +343,10 @@ func (co *connection) close() {
 
 // lockingAppend handles a single append request on a given connection.
 func (co *connection) lockingAppend(pw *pendingWrite) error {
+	// If connection context is expired, error.
+	if err := co.ctx.Err(); err != nil {
+		return err
+	}
 	// Don't both calling/retrying if this append's context is already expired.
 	if err := pw.reqCtx.Err(); err != nil {
 		return err