Fix intermittent integration test failures

[rwood-moz]

What changed?

Fix some mailstrom intermittent integration test failures.

Why?

The tests are mostly (98%) stable but would be nice to get rid of the intermittent failures.

Limitations and Notes

Changes made by Codex AI; we will see how these changes prove out after a couple weeks in CI. More details:

Changes made so far:

• Fixed IMAP deletion test calls in test_messaging.py and test_messaging.py so single message ids are passed as lists to imap.delete_messages(). This prevents multi-digit ids like "25" from being iterated character-by-character.

• Hardened IMAP.delete_messages() in IMAP.py so it verifies \Deleted by fetching message flags after STORE, rather than depending on the exact shape of the STORE response data.

• Improved JMAP.wait_for_message_to_arrive() in JMAP.py to dynamically resolve the active account’s Inbox id, poll immediately, wait up to 7 minutes, accept one or more matching messages, and log the account/inbox being checked.

• Updated JMAP.query_email() in JMAP.py to make collapse_threads configurable, preserving the existing default while allowing arrival polling to query with collapse_threads=False. This should help self-send CC/BCC tests where collapsed thread results may hide the inbox copy.

Applicable Issues

Fixes #230.

QA Log

• Ran the integration tests on my machine against stage several times, and
• Ran the nightly mailstrom integration tests GHA job against this branch in CI several times and didn't see any intermittent failures thus far.

[aatchison]

Review: Approve with nits ✅

Ran a multi-lens review of this change. TL;DR: this is a well-scoped de-flaking PR that fixes real root causes rather than loosening tolerances — the key property for a de-flaking change is that a genuinely-broken message/arrival still fails loudly, and that holds here. Nothing blocks merge.

What's good

• delete_messages([test_msg]) — stops a bare string id (e.g. "25") from being iterated character-by-character, which silently flagged unrelated messages \Deleted while the test still passed by re-finding the intact subject. Genuine intermittent-corruption source and a likely mechanic behind A couple of mailstrom integration test intermittent failures #230.
• Re-fetching FLAGS instead of asserting on data[0] (the optional STORE untagged echo, whose shape/timing varies by server) verifies authoritative server state on the same still-SELECTED connection — API-correct and de-flakes without weakening the assertion.
• Resolved inbox_id replacing hardcoded inMailbox='a', guarded by asserts — turns a slow timeout-to-0 into an immediate, diagnosable failure.
• Loop math is exact: (7*60//5)+1 = 85 checks with the if checks > 1 sleep guard → 84 × 5s = exactly 420s, no overshoot, and an already-delivered message returns ~5s faster.
• query_email signature change is fully backward-compatible — collapse_threads defaults to the old hardcoded True; all other callers unaffected.

Non-blocking suggestions

1. (low) IMAP.delete_messages still has the footgun. Only the call sites were fixed; the helper still iterates its arg, so a future caller passing a scalar id re-opens the exact char-by-char wrong-deletion bug. Consider hardening the helper:

   def delete_messages(self, msg_ids):
       if isinstance(msg_ids, str):
           msg_ids = [msg_ids]
       ...

   and tightening the docstring to state it takes a list of ids.
2. (nit) # = 85 comment. It restates a computed value and will lie if the timeout is tuned. The load-bearing detail is why the +1 exists — the first iteration skips its sleep, so +1 keeps total sleep == max_wait_seconds (420s). Worth documenting that coupling near the if checks > 1 guard instead of the literal 85.
3. (nit) len(found_email_ids) >= 1 is just bool(...); if found_email_ids: reads clearer now that the intent is "any match found."

Out of scope (pre-existing, possible follow-up)

• get_mailbox_by_name raises UnboundLocalError on its swallowed-exception path (results never assigned before the assert).
• wait_for_message_to_arrive returns 0 on timeout despite the docstring saying -1.

Neither is introduced by this PR.

[aatchison]

approved with nits!

[rwood-moz]

approved with nits!

Thanks @aatchison I made some of the suggested updates and now it needs a re-approval please, thanks