Improve error messaging for non-parsable datasets

[justinormont]

I'd recommend improving the current error message:

machinelearning/src/Microsoft.ML.AutoML/ColumnInference/ColumnInferenceApi.cs

Lines 120 to 123 in e5a19af

	if (!splitInference.IsSuccess)
	{
	throw new InferenceException(InferenceExceptionType.ColumnSplit, "Unable to split the file provided into multiple, consistent columns.");
	}

It currently says, "Unable to split the file provided into multiple, consistent columns.", which is rather uninformative and non-actionable.

Perhaps, as I think @briacht is suggesting, have it list the acceptable file formats we can parse: "Unable to split the file provided into multiple, consistent columns. Readable formats include delimited files such as CSV/TSV. Check for a consistent number of columns and proper escaping and quoting.".

This messaging now includes, the problem, and next steps for the user.

I mention delimited as AutoML supports more than CSV/TSV as it tries tab, comma, space, semi-colon as the separator (src). If we run into other common separators, we can trivially augment this list. One candidate is the vertical bar |.

[justinormont]

Transferring to main ML․NET repo as the PR to fix will be for that repo.