I can't explain how/when it happened, but you are right that the gaia_dr2.light_curves table is short for some reason. I've reloaded it from the origin FITS images downloaded shortly after the data release and the table now contains 548,320 unique source_id values which is still short of the number of variable sources you quote. If I look at the CSV files of light curves downloaded at the same time, there are 549,371 uniq source_ids -- again short of the number you mention but also different than the number derived from the FITS files.
FITS files were not available for all tables just after the data release and so some tables were loaded from CSV files, my guess would be that the download files were still being generated/modified at the time we got them, however I cannot say whether files obtained today would be different yet again. The specific source_id you mention *is* now in the table because of the reload.
Lastly, different data centers may have constructed/augmented the Gaia tables in different ways. The Data Lab tables were built from what was available as a data download, not as an explicit mirror of ESA or other databases. If there is something missing that you need we will try to provide it. Please let us know if you find problems with any of the other tables.