Introduce int32 and int64 types in order to easier transform flow entries into parquet/avro types

[stloyd]

we might need to introduce here int32 and int64 in order to easier transform flow entries into parquet/avro types

Originally posted by @norberttech in #770 (comment)

[norberttech]

I'm putting this on hold for now. I still think it's necessary to introduce this int32/int64, but I don't want to complicate current developer experience without strong and solid use cases.

[norberttech]

One quick idea is that for formats like parquet/avro where type length is important, we can use schema metadata.
In most cases, the schema is inferred from the dataset but when passed explicitly it could also carry additional metadata which could help us to choose a better destination type.