fix: do not miss new messages while expunging the folder

[link2xt]

This should fix flaky test_verified_group_vs_delete_server_after.

Fix #5201

[link2xt]

Finally non-flaky CI, but need to think of a proper fix now.

[link2xt]

On Windows tests failed once with "Backup provider did not start in time" message, but this is likely just because timeout is 10 seconds and GitHub Actions seem to sometimes pause/stop VMs:
https://github.com/deltachat/deltachat-core-rust/actions/runs/9374467861/job/25810847989?pr=5669
Looks like another thing we may want to fix to get non-flaky CI, but not caused by changes in this PR.
Otherwise CI is much less flaky with this change.

[iequidoo]

Do i correctly understand that this is just in case and not related to the fix itself? Can it be done under if self.server_sent_unsolicited_exists(context)? if so?

[link2xt]

This is not just in case, without this line tests will fail.

If IDLE exits because of NewData, we want session.new_mail be true next time fetch_new_messages is called. Previously fetch_new_messages called session.select_with_uidvalidity and used its return value to decide if it should try to fetch messages. session.select_with_uidvalidity returned true if the folder was already selected, so most of the time fetch happened.

In cases other than NewData it may be not necessary, but other cases are rare so better try fetching. E.g. in case of manual interrupt it seems better to fetch, because who knows what kind of broken IMAP servers are there, closing and reopening the app should not rely on correctly working IDLE to try fetching messages.

I think IDLE response does not get into unsolicited channel, so checking for unsolicited EXISTS is not enough. Unsolicited EXISTS is an EXISTS response that you get e.g. when issuing a FETCH or STORE command.