Fix notifier rc handling and spurious 'session finalized as incomplete' WARN

[hostarts]

Summary

Two related log-noise / correctness issues observed on every successful run, both visible in the same trailing log block:

[INFO] [main] Sending email report to ...
Skipping email report (EMAIL_ON_SUCCESS=no)
[WARN] [main] Failed to send email report (backup data preserved)
[INFO] [main] Slack notification sent
[INFO] [main] Session ended successfully - exit code 0
[WARN] [cleanup_on_exit] SQLite session finalized as 'incomplete' (exit code 0)

Neither WARN reflects an actual problem.

Bug 1 — Intentional notifier skip logged as failure

send_backup_report() distinguishes:

• 0 — delivered
• 1 — transport failure
• 2 — intentionally skipped (module disabled, EMAIL_ON_SUCCESS=no, EMAIL_ON_FAILURE=no)

But all four call sites used the binary form if send_backup_report ...; then INFO; else WARN; fi, collapsing rc=2 into the failure log line. Operators running with EMAIL_ON_SUCCESS=no (very common — "only email me when something breaks") get a misleading WARN on every successful run.

This PR adds a small _handle_notifier_rc() helper that interprets the three return codes correctly and is wired into all eight call sites (4× email, 4× Slack from the sister Slack PR).

Side benefit: _EMAIL_SENT / _SLACK_SENT sent-guards are also set on intentional skip, so later code paths (handle_sigterm, cleanup_on_exit) don't try to re-send a notification the operator explicitly suppressed.

Bug 2 — cleanup_on_exit overwrites successful session log line

The catch-all sqlite_session_end block in cleanup_on_exit() exists to finalize sessions on early-error paths and signal exits. On a normal successful exit, main() has already called sqlite_session_end with status='success', and _SQLITE_SESSION_ENDED is set. The idempotency guard inside sqlite_session_end() correctly drops the duplicate DB write — but the surrounding vmbackup.sh log line still emits:

[WARN] [cleanup_on_exit] SQLite session finalized as 'incomplete' (exit code 0)

…which suggests the session was downgraded when in fact the DB is fine. Fix: gate the catch-all block on _SQLITE_SESSION_ENDED != 1.

What changed

• vmbackup.sh:
  • New _handle_notifier_rc() helper (24 lines, defined just above cleanup_on_exit).
  • All 8 notifier call sites converted from if send_X; then INFO; else WARN; fi to send_X || _rc=$?; _handle_notifier_rc ....
  • cleanup_on_exit's SQLite finalize block gated on _SQLITE_SESSION_ENDED != 1 (avoid spurious WARN; existing idempotency guard inside sqlite_session_end() is unchanged and continues to protect data).
• CHANGELOG.md: [Unreleased] entries under ### Fixed.

Dependency

Touches both the email and Slack call sites. Based on feat/slack-notifications (#3); please merge #3 first or merge both together.

Test plan

• Helper return-code matrix (rc=0,1,2,3) verified manually: rc=0→INFO+SENT, rc=2→DEBUG+SENT, rc=1/other→WARN+!SENT.
• bash -n vmbackup.sh passes.
• Real backup run on a production host: previous "Failed to send email report" + "finalized as 'incomplete'" WARNs no longer fire on a successful single-VM run with EMAIL_ON_SUCCESS=no.

🤖 Generated with Claude Code