Improve error experience when LoadCsv throws

[eerhardt]

If you have .csv file that has some error in the data, for example:

Date | Daily minimum temperatures
1/1/1981 | 20.7
1/2/1981 | 17.9
1/3/1981 | 18.8
1/4/1981 | 14.6
1/5/1981 | 15.8
...
7/20/1982 | ?0.2
7/21/1982 | ?0.8

Here you get an exception:

System.FormatException: Input string was not in a correct format.
   at System.Number.ThrowOverflowOrFormatException(ParsingStatus status, TypeCode type)
   at System.String.System.IConvertible.ToSingle(IFormatProvider provider)
   at System.Convert.ChangeType(Object value, Type conversionType, IFormatProvider provider)
   at Microsoft.Data.Analysis.DataFrame.Append(IEnumerable`1 row, Boolean inPlace)
   at Microsoft.Data.Analysis.DataFrame.LoadCsv(Stream csvStream, Char separator, Boolean header, String[] columnNames, Type[] dataTypes, Int64 numberOfRowsToRead, Int32 guessRows, Boolean addIndexColumn, Encoding encoding)

Which isn't super helpful to tell you where or what was causing the problem. We should log a more helpful error in this situation so users know what is wrong.

We should also consider having a mode where things like this turn into N/A values, like Single.Nan.

[chriss2401]

I've wrapped a simple try catch around the function where it "blows up" :
https://github.com/chriss2401/corefxlab/commit/e1f74ec49ffaf3803052b28c4ff20c5f507ad63a

Regarding the NaN comment, would it make sense that it by default converts it to Single.NaN inside the catch area, doesn't throw an exception but informs the user that this value was converted to NaN ?

[chriss2401]

@eerhardt @pgovind could this fix potentially also go into dotnet/corefxlab#2927 ?

[pgovind]

Yup, @chriss2401, I think it makes sense for this fix to go into dotnet/corefxlab#2927. At some point, I expect to remove the LoadCsv method from DataFrame and move it to DataFrameIO.

I think the Single.NaN suggestion makes sense, however, I think we'd have to check column.DataType for each type and specialize for the Single/Double cases? It might be easier to just make the value null in the catch block and delete the following if (value is null) block on line 495