[Data Transfer] elasticc schema varies per partition

[JulienPeloton]

Even if the Spark DataFrame has a unique schema, when polling the data different partitions can get different schemas. This happens if a column in a partition has only null values -- in which case you will get e.g.:

> 10:50
124,125c125,126
< cats_broad_class: double
< cats_broad_max_prob: double
---
> cats_broad_class: null
> cats_broad_max_prob: null

[JulienPeloton]

Done in #166