[2.x] fix: lower SMTP ping threshold to 20s to survive idle-disconnect from relays

[imorland]

Summary

Symfony's SmtpTransport reuses the SMTP connection across sends within a process and only NOOPs the connection when the inter-message gap exceeds pingThreshold (default 100s). AWS SES has been observed closing idle connections in the 30–90s range, so messages spaced 30–100s apart get written into a dead socket and the relay responds with:

451 4.4.2 Timeout waiting for data from client

The reporter's data (failures clustered late in the lifetime of each long-lived queue worker, never near the start) clearly fingerprints connection-age dependence.

Set pingThreshold to 20s on the Symfony transport in SmtpDriver::buildTransport() — well under any reasonable relay idle timeout, with negligible NOOP overhead.

Fixes #4614

Changes

• framework/core/src/Mail/SmtpDriver.php — after building the transport via the factory, call $transport->setPingThreshold(20) (guarded by instanceof SmtpTransport so the cast is safe even though EsmtpTransportFactory always produces SMTP transports).

Test plan

• Manual: configure SMTP against AWS SES; run a long-lived queue:work with a moderate trickle of email-bound notifications; confirm the 451 4.4.2 Timeout errors no longer appear in storage/logs/flarum-*.log.
• Existing mailer integration tests pass.

No automated test for this — the bug only manifests with a real long-lived SMTP connection against a remote server with idle-disconnect behavior. Reproducing in a unit test would require simulating the transport's connection lifecycle, which is heavyweight relative to the fix.

Notes

• 20s was chosen as a balance: well below any observed relay idle timeout (SES's lower bound is ~30s in the field), but high enough that NOOP overhead is rare. Could be made configurable via a mail_smtp_ping_threshold setting in a future PR if any operator runs into a relay needing different tuning, but the fixed default handles the common cases (SES, Mailgun, Postmark, Gmail SMTP) without ceremony.