Backport: Move vuln data source mirroring off JDO - #7384

Refactors the `mirror-vuln-data-source` activity to use JDBI instead of JDO. This enables proper use of batch operations which significantly slashes database round trips and consequently wall clock time.

Having tested this with OSV's Ubuntu ecosystem, the new implementation issues over 1000x fewer database queries.

The old implementation also showed drastic degradation after ~30k synchronized vulnerabilities. I was able to observe throughput dropping from ~210 vulns per second to ~11/s. This was caused by queries on the `VULNERABLESOFTWARE` table that happened to match large numbers of records, e.g. those targeting the `linux` package. These queries were executed once per affected version. And naturally during a mirror run, the data already present in the DB grows, making the slowdown worse as the run progressed. This is likely the root cause of DependencyTrack#5934.

With the refactored code, processing the OSV Ubuntu ecosystem locally is stable at ~280 vulns per second. For Chainguard it's ~3200/s. The difference between the two is caused by Ubuntu publishing few(-er) advisories with *many* affected versions, and Chainguard publishing *many* advisories.

Note that the old JDO-based logic was not removed and is still used by a few REST API v1 endpoints. I opted to not do a full replacement now to keep the blast radius constrained.

Adapted for 5.1.x, where AFFECTEDVERSIONATTRIBUTION.LAST_SEEN still exists
and is NOT NULL: the column is written alongside FIRST_SEEN.

Signed-off-by: nscuro <nscuro@protonmail.com>
(cherry picked from commit 58b679f)