quic batched send metrics are unreliable on error

[t-nelson]

Problem

rpc sts only increments its send failure once on error, regardless of batch size in the case of batched sends

agave/send-transaction-service/src/send_transaction_service.rs

Lines 748 to 760 in b27c80a

	let result = if wire_transactions.len() == 1 {
	Self::send_transaction(tpu_address, wire_transactions[0], connection_cache)
	} else {
	Self::send_transactions_with_metrics(tpu_address, wire_transactions, connection_cache)
	};
	if let Err(err) = result {
	warn!(
	"Failed to send transaction transaction to {}: {:?}",
	tpu_address, err
	);
	stats.send_failure_count.fetch_add(1, Ordering::Relaxed);
	}

it can't really do any better because the quic tpu client batched send implementation bails on the first error, discarding statuses of any prior and pending sends

agave/quic-client/src/nonblocking/quic_client.rs

Lines 506 to 512 in cc3afa5

	for f in futures {
	f.await
	.into_iter()
	.try_for_each(|res| res)
	.map_err(Into::<ClientErrorKind>::into)?;
	}
	Ok(())

Proposed Solution

investigate whether waiting to collect all results of a batched send has significant performance penalty.
if so, split rpc sts metrics to report failed single and batched sends separately.
if not, rework the quic client's batched send such that in can return the error count.