This doesn't check any content of the PDF, but ensures that running
typst does not produce an error.

A lot of time in test_browse was spent retrieving path points and
computing metadata. Why not save it while we still have the path ready?

It turns out that adding multiple thousands of track points via the ORM
is quite slow, which is especially noticable in the browse_paged test
(as 50 tracks are added there).

This provides the fast_set_path method, which goes directly to the
database, utilizing the executemany() capability.

On the CI, the gpx-influx decreased test runtime from ~4min to ~18min.
This change should make it much faster again.

With the values being computed ourselves, they differ a bit from the
previous gpxpy values. Therefore, we adjust the expected values in the
tests.