AlgoScan

Existing tests mocked the queryServerIf transport, leaving qsi.queryServer -
the real UDP/TCP behavior - uncovered. Add tests against a loopback DNS
server for: retrying over TCP after a UDP error, retrying over TCP on a
truncated UDP response, advertising the 1232-byte EDNS0 buffer, and
surfacing the per-server error when all attempts fail.

Relay SRV resolution failures were logged only on testnet and devnet; on
mainnet the error was dropped and the only trace was an INFO "got no relay
DNS addrs" line. A mainnet node that resolves no relays then runs with an
empty phonebook, connects to zero peers, and reports itself caught up at a
stale round, with no warning an operator can act on.

Log the relay resolution error as a warning on every network and raise the
empty-relay case to a WARN stating the operational impact. Archive lookups
stay informational, since archival peers are optional for most nodes.

When a query failed against every configured server, query() discarded each
server's error and returned only "no answer for (name, type) from DNS
servers [...]", which is indistinguishable from a missing record. The cause
(timeout, truncation, blocked TCP, ...) was lost, leaving operators with a
lookup that fails here while dig resolves it and nothing to act on.

Join the per-server errors onto the returned error so the real cause is
visible.

queryServer only retried over TCP when a UDP response came back truncated
(TC bit set). If the UDP query itself errored - e.g. the response was
dropped or UDP/53 is filtered - it returned immediately without trying TCP.

Fall through to the TCP attempt on a UDP error too, giving up only once the
context is done. Set Client.Timeout as well so dial, write and read are
bounded by the configured DNS timeout instead of miekg's 2s default.