Bug in atc insertion/oversampling of lmdb

[Aastedet]

Currently, A10BJ06 (semaglutide) seems to be the only atc code being inserted/oversampled. So about 30% of all rows of the synthetic lmdb data is value:

summary(factor(register_data$lmdb$atc))

Since we stopped including based on GLP1-RAs (atc ^A10BJ), we don't "need" to oversample these so much anymore, since we don't need to cover edge cases within them based on vnr/name.

Relates to #146, where we also have to update the functions generating the fake lmdb data.