These primary keys were converted to bigints in migrations, but the
sequences were left unmentioned. If the original migrations are run on
postgresql 10.0+, then this leads to a mismatch in column types vs sequence
types. This migration fixes these mismatches.
If the original migrations were run on postgresql < 10, all sequences were
bigints anyway, and this migration is a no-op.
If the sequence is a bigint, then postgresql doesn't output that fact in the
statement dump.