When we introduced timeout for snmp in ZBX-4393 we also limited library's retries to 0.
There was a discussion and I'm still sure it was reasonable to limit library's retries (default was =5) as with server's timeout (which is often =30 in production) it would lead to big delays.

But after ZBXNEXT-98 in 2.2.3 and discovered and investigated ZBX-8528 (see more details in this issue namely) I suggest and ask to allow single retry on library level until we will not introduce ZBXNEXT-1096.

In any case it will have more positive effect than negative.
I spent enough time to weigh it

Even decreased rate of "first network error/recovering" because of random lost UDP packets in zabbix log is an improvement.