update-citations workflow discards the script's diagnostics on timeout

Spun out of #3619 so the finding isn't lost when that issue closes as `wontfix`.

`.github/workflows/update-citations.yml:52` runs:

```yaml
timeout 90 python bin/update_scholar_citations.py
```

Python buffers stdout when it isn't attached to a TTY, so when `timeout` kills the process at 90s the buffer is discarded and the log shows nothing — not even the `Fetching citations for Google Scholar ID ...` line the script prints on startup. The run looks like it did nothing, when in fact it was mid-request.

@dfuchss verified on #3619 that running with `python -u` restores that output.

**Fix:** add `-u`.

```diff
- timeout 90 python bin/update_scholar_citations.py
+ timeout 90 python -u bin/update_scholar_citations.py
```

This does **not** fix the underlying timeout — Google Scholar rate-blocks GitHub-hosted runner IPs, which is why #3619 is `wontfix` and why a longer timeout doesn't help. It only means the next person to look at a failed run can see how far it got before being blocked, instead of an empty log.

Open .github/workflows/update-citations.yml at line 52 and inspect the invocation of bin/update_scholar_citations.py. Check the workflow command's Python output behavior when timeout stops the process. Done means the command uses unbuffered output so a timed-out run retains the script's diagnostic lines in the GitHub Actions log.

Vom Indexierungsmodell aus dem Issue-Text verfasst.