iterate: avoid turning off qname minimization in a case

Fixes https://gitlab.labs.nic.cz/knot/knot-resolver/issues/339.

Lots of tests done during today, and it seems good, but I'm doing some more.

So far it seems good to me. Respdiff runs, alternating between this PR and master:

• 8+8 of iter.udp.j128 – there it mainly seemed the variability of results decreased by this PR significantly (mean deviation fell to a fraction and that wasn't caused just by a couple of runs),
• 5+6 of fwd-udp-unbound.udp.j128 – nothing significant observed (deviation is relatively low there),
• 4+5 of fwd-tls-kresd.udp.j128, nothing significant observed (these tests have higher variability in timeouts).

mentioned in issue #339 (closed)

There was not additional feedback from RIPE 76 so let's merge this and see.

mentioned in commit cdbcb1ea

merged