"resolver" test fails intermittently

The recently-added (only on main) check for a SERVFAIL response to a TCP query with an empty question section fails pretty frequently. The check was added in commit 2f3ded76 (part of !5616 (merged)). Despite a code comment suggesting otherwise, the test fails due to dig hitting a timeout instead of receiving a SERVFAIL response:

$ cat dig.ns5.out.70

; <<>> DiG 9.17.21 <<>> -p 29349 @10.53.0.5 -b 10.53.0.5 +tcp tcpalso.no-questions. a +tries=3 +time=4
; (1 server found)
;; global options: +cmd
;; connection timed out; no servers could be reached

Example occurrences:

• https://gitlab.isc.org/isc-projects/bind9/-/jobs/2197750
• https://gitlab.isc.org/isc-projects/bind9/-/jobs/2197749
• https://gitlab.isc.org/isc-projects/bind9/-/jobs/2197279
• https://gitlab.isc.org/isc-projects/bind9/-/jobs/2196581
• https://gitlab.isc.org/isc-projects/bind9/-/jobs/2185707
• https://gitlab.isc.org/isc-projects/bind9/-/jobs/2185704
• https://gitlab.isc.org/isc-projects/bind9/-/jobs/2185288

Most of these happened on FreeBSD, but two happened on Linux (under TSAN), so the common denominator seems to be "machine under load" rather than "only happens on FreeBSD".