Named CSV instance eats errors

[MichaelXavier]

I notice that the instance for CSV eats errors on parse. It first parses rows into NamedRecord and then parses to an Either String a, discards the error message and skips the row. I found this surprising as default behavior and it does not appear to be documented. In my case, If you parse a CSV file with some invalid rows, it simply skips them and returns a successful result with the rows that worked. If you have a missing column, it parses successfully and returns that there were no rows. The MonadThrow requirement for IntoCSV seems to be unused.

I could see a few ways forward:

1. This is undesirable default behavior and the Named instance will throw a new exception that wraps the String error message. We could create a newtype like Named called NamedSkipErrors or something more pithy that uses this current implementation of eating errors.
2. This is valid default behavior and we don't want to break existing installs, so we create a NamedThrowErrors or something more pithy that throws on invalid rows.
3. This could be added to either of the above as newtype NamedOrError a = NamedOrError (Either String a) which also never throws but captures per-row errors and lifts them into the types so they can be dealt with.

Because these are all solved by newtypes, this is something I can do myself in userland code without changing the library in the mean time.