Split data ingestion into process and validate

[pipliggins]

Most of the time spent on the ingestion pipeline is taken up by the process of validating the fhirflat format, which requires creating a pydantic class object for each resource then re-creating the FHIRflat parquet.

Splitting up the pipeline into two steps (which can be run together using a single command if desired) should allow for greater parallelisation (see #44) and allows for data to be read in and edited before validation, without having to run the whole pipeline again for a few validation errors.