KeyToValueMapping API is inconsistent with rest

[TomFinley]

Pursuant to #2064 and the work done in #2239, column mapping transforms are to have an output and input column. However, I notice something like this:

machinelearning/src/Microsoft.ML.Data/Transforms/ConversionsExtensionsCatalog.cs

Lines 72 to 73 in 578c188

	public static KeyToValueMappingEstimator MapKeyToValue(this TransformsCatalog.ConversionTransforms catalog, string inputColumnName)
	=> new KeyToValueMappingEstimator(CatalogUtils.GetEnvironment(catalog), inputColumnName);

This has just one column specified, that is, there is no convenient way to map an input to an output column and have the mapping have a distinct names, without working through the params overload which is a bit less convenient. This is inconsistent with the direction summarized by @sfilipi in #2064 as well as some of our other APIs. For example, in this same file:

machinelearning/src/Microsoft.ML.Data/Transforms/ConversionsExtensionsCatalog.cs

Lines 30 to 31 in 578c188

	public static HashingEstimator Hash(this TransformsCatalog.ConversionTransforms catalog, string outputColumnName, string inputColumnName = null,
	int hashBits = HashDefaults.HashBits, int invertHash = HashDefaults.InvertHash)

machinelearning/src/Microsoft.ML.Data/Transforms/ConversionsExtensionsCatalog.cs

Lines 49 to 50 in 578c188

	public static TypeConvertingEstimator ConvertType(this TransformsCatalog.ConversionTransforms catalog, string outputColumnName, string inputColumnName = null,
	DataKind outputKind = ConvertDefaults.DefaultOutputKind)

This method should be similar to the others in terms of the string outputColumnName, string inputColumnName = null signature. Note that this mistake extends down into the associated estimator and transformer constructors as well, so it will have to be fixed there as well.

/cc @sfilipi