Fix #7150: passwd-verify.lua wipes auth cache during any nginx outage

[wryfi]

Contribution Guidelines

• I've read the contribution guidelines and wholeheartedly agree them

What does this PR include?

Short Description

Fixes #7150.

passwd-verify.lua wiped the Dovecot auth cache for affected users whenever
a host reboot, a full-stack snapshot/backup, or any transient Docker network blip.

On a network-level failure, https.request returns nil plus an error string
(e.g. "host or service not provided, or not known"), so c is not a numeric HTTP
status code. The existing if c ~= 200 and c ~= 401 check treated this identically
to a credential rejection and returned PASSDB_RESULT_PASSWORD_MISMATCH, which clears
the cache entry — causing hard auth failures for reconnecting clients (iOS Mail in
particular, which has no Keychain fallback).

This PR adds a guard ahead of the HTTP-status check: when c is not a number, the
request failed at the network level, so it returns PASSDB_RESULT_INTERNAL_FAILURE
("Upstream unreachable"), which keeps the existing cache entry instead of wiping
it. Genuine HTTP responses (200/401, and 4xx/5xx errors) are handled exactly as before.
The log concatenation is also hardened to tostring(c) so it can't error on a
non-string/nil value.

This restores the pre-2025-05 behavior where a brief nginx interruption was benign for
already-authenticated clients.

Affected Containers

• dovecot-mailcow

Did you run tests?

Yes — tested on a live mailcow instance by overriding passwd-verify.lua and exercising
the real-world scenario that originally surfaced the bug.

What did you tested?

The Dovecot auth path (dovecot-mailcow) through a full-stack restart, using my backup
procedure as the trigger: stop the entire stack, snapshot the volumes, then start the
stack again. This is exactly the kind of operation that leaves nginx:9082 unreachable
while Dovecot auth workers are still serving reconnecting IMAP clients.

I first noticed the failure on the stock file, then applied the patched
passwd-verify.lua as an override and repeated the same backup/restart cycle with active
IMAP clients connected.

What were the final results? (Awaited, got)

• Stock file, full-stack restart → awaited: the documented bug. got: confirmed —
  auth cache entries for connected users were wiped and clients hit hard authentication
  failures during the nginx-unavailable window.
• Patched file, same full-stack restart → awaited: cache entries retained, clients
  unaffected. got: as awaited — no cache wipe, and connected IMAP clients stayed
  authenticated across the entire stop/snapshot/restart cycle, with no re-auth prompts.
• luac -p data/conf/dovecot/auth/passwd-verify.lua → awaited: no syntax errors.
  got: syntax OK.