Use PooledVector for "categorical" columns like "bus type code" (buses.ide)?

[nickrobinson251]

Bus type code is either 1, 2, 3 or 4. Currently this is parsed into a Vector{Int64}. But this could potentially be more efficient in a couple ways:

1. the integers could be parsed into an Int8 rather than Int64 (technically i suppose we only need 2 bits, but Int8 is probably the smallest type it is actually practical for users to be given).
2. rather than being stored in an N-length Vector (with N T=Int64 integer values), it could be stored as a PooledVector (with N UInt8 values and a 4-element UInt8 => T values). And both options could be combined (e.g. pool and have T=Int8).

Option 2 doesn't really sound worth it on storage-efficiency alone, but it could be worth it (i.e. provide practical performance improvements to users) depending on how the pooled columns (e.g. "bus type code") are going to be used, because certain operations (joins, mapping, ...) can be very efficient on PooledVectors (as they can work with the 4 pooled values, rather than all N entries).

So we should probably do 1 (i.e. change to Int8s) and then investigate how these columns will be used to decide about 2 (pooling).