Fix handling of GSAK multi DB backups

[nagisml]

see #946

GSAK import: support multi-database backups

A GSAK backup .zip holds one sqlite.db3 per database (plus gsak.db3 with the saved filters). Until now the import took only the first database and unpacked the whole backup to %TEMP% without removing it.

• Lists every database in the backup (without unpacking it); the user picks which to import. Each target is named after its GSAK folder (editable) and is created if it doesn't exist.
• Existing OpenSAK databases are only touched after confirmation: Overwrite / Merge / Skip.
• Name clashes with stray files on disk are detected before the import starts (a free name like Default-2 is proposed instead).
• Imports one database at a time: unpack → import → delete the temp copy. A failing database is logged and skipped rather than aborting the run.
• If gsak.db3 is present, offers the existing GSAK filter import afterwards.
• A single .db3 file or a zip with one database works as before (a one-row list).
• Fix: load_gsak_filters() never closed gsak.db3, which kept the file locked on Windows.

Tested on a real 23-database / 9 GB backup; unit tests and mypy pass.

[AgreeDK]

@nagisml
Thanks for this, it's a big improvement for everyone migrating from GSAK. This fixes #946.

I've tested it locally, both on its own and merged on top of beta: unit tests and mypy are green.

What I like:

• Only one database is unpacked at a time, and the copy is removed afterwards. Before, extractall() copied the whole (possibly multi-GB) backup to the temp folder.
• extract_gsak_member() always writes to dest_dir / basename, so a crafted member name can't escape the folder.
• Skip is the default button when target databases already exist, not Overwrite.
• clear_opensak_cache_data() covers all five tables with a foreign key to caches.
• closing() around the sqlite connection fixes the Windows file lock.

A few things to consider (none of them blocking):

1. Overwrite can't be undone. It's confirmed by a dialog, so it's acceptable for now. But it's a good example of why Backup is now PC - DT Filter is not displaying correctly #1 on the roadmap, and I'll keep it in mind when we design that.
2. The worker switches the global database engine with init_db() in a background thread during the import. That pattern already existed on beta, so it's not new in this PR, but it now runs many times in a row for a multi-database import. The dialog is modal, so the risk is low. Just flagging it as a known weak spot.
3. The distance fix in _switch_database recalculates synchronously on the GUI thread when distances aren't up to date. On a large database the UI may freeze briefly on the first switch after an import. Fine for now, but maybe a candidate for the background worker later.

Needed before merge: this PR conflicts with #948 in tests/e2e-tests/test_e2e_filter.py (all three of your PRs fix the same flaky test). I'll merge #948 first, since it reuses the existing wait_for_refresh() helper. Could you then rebase on beta and drop the change to that file?

I'll put fixes #946 in the merge commit message, so the issue closes when this reaches main.

[AgreeDK]

@nagisml
Update: no need to rebase this one. I'll resolve the conflict in test_e2e_filter.py myself so we can get it merged. I'll keep the version from beta (from #948).