Use read_csv_to_dataframe in CSV validation

[emlys]

Validation has its own logic for reading and checking CSVs. This is now redundant to utils.read_csv_to_dataframe and less comprehensive. It would make sense to read CSVs for validation in the same way that we're going to read them in execute.

Currently, CSV validation check that:

• the csv exists
• the file has the necessary permissions
• exact column names exist in the csv

With utils.read_csv_to_dataframe, we can also easily check that:

• at least one column name matching a pattern exists in the csv
• values can be cast to the appropriate data type for each column

related issues

This makes sense to do before #327 because read_csv_to_dataframe handles path expansion.

I'd also like to revisit whether we can drop support for Excel files (which is only allowed in one table in HRA) rather than writing additional code to support them.