Deal with local recv() truncation.

In the case where an upstream cache sends a UDP response that would overflow
the djb cache's default receive buffer, then djbdns would treat this as an
invalid response.  The norm nowadays is the send >512b UDP responses,
especially for TXT RRs.  It looks like up to around 4KB is deemed acceptable in
most cases I've investigated.

So, in the case where we locally end up reciving a truncated packet by way of
recv() because the local UDP buffer is too small, treat that like the TC bit
was set, because really we can know the response was truncated.

Therefor check the irrelevant (inappropriate response) data first, then if the
buffer was fully received (it might be that the response fits exactly, but
short of parsing this buffer there is no simple way to confirm this, so just
assume it's unlikely to get an exact sized buffer back and retry using TCP
anyway).  Yes, this is a waste of resources in this specific case, but so be
it.

Signed-off-by: <jaco@uls.co.za>

--- djbdns-1.05.o/dns_transmit.c	2024-05-27 13:20:25.788463090 +0200
+++ djbdns-1.05/dns_transmit.c	2024-05-27 14:13:38.786335627 +0200
@@ -266,9 +266,9 @@
       if (errno == error_connrefused) if (d->udploop == 2) return 0;
       return nextudp(d);
     }
-    if ((size_t)r + 1 > sizeof udpbuf) return 0;
 
     if (irrelevant(d,udpbuf,r)) return 0;
+    if ((size_t)r + 1 > sizeof udpbuf) return firsttcp(d); /* if udp overflowed, retry with TCP */
     if (serverwantstcp(udpbuf,r)) return firsttcp(d);
     if (serverfailed(udpbuf,r)) {
       if (d->udploop == 2) return 0;