Since you are not telling it to run more that one job at a time (by default n_jobs = 1), your easiest option would be to specify a reduced number of reads per batch. You can do this using --p-reads-per-batch when you call classify-sklearn. For n_jobs = 1, the default reads per batch is either 20,000 or the total number of sequences, whichever is smaller.
For example, if you want to only run 5,000 sequences per batch, your input would look like:
qiime feature-classifier classify-sklearn \
--i-classifier /home/qiime2/silva-138-99-515-806-nb-classifier.qza \
--i-reads rep-seqs-dada2.qza \
--o-classification taxonomy.qza \
--p-reads-per-batch 5000
You may have to play around with this number to get it to work. Alternatively, if you had access to a computer with more memory than the one you are currently using, you could try running classify-sklearn on it.
Hope this helps!