pass arguments to read_csv

[alexwalkerepi]

When using bq.cached_read on the query below, the subpara column is initially read as a str, but when reading from the cache, it's read as an int. Ideally, we'd be able to pass arguments like:
dtype={'subpara': str} to pd.read-csv.

SELECT
  DISTINCT SUBSTR(bnf_code, 1, 7) AS subpara,
FROM
  ebmdatalab.hscic.normalised_prescribing_standard AS prescribing

[evansd]

One option would be to muck about with the CSV headers and encode the type in them (eg. we'd store str:subpara rather than just subpara) and then we'd strip the types out and use them when we read the cached file back. I don't know how easy Pandas makes that but I'm sure it's possible.

[evansd]

It also looks like Pandas supports comments in CSV files (at least to the extent of ignoring them on read) so we could write the types into an initial comment line in the file if that's easier than modifying the headers.

[sebbacon]

Or we could store the types in the fingerprint_path file. But that would require a bigger refactoring. I like the comments idea.

[alexwalkerepi]

This is definitely a more robust solution than passing arguments to read_csv. It took me a while to realise that the types had changed between the query and the cache. This way it should keep the same types by default.