Improve error handling for GaussianCopulaSynthesizer: numerical_distributions

[npatki]

Problem Description

In SDV 1.0, I will be able to specify numerical distributions using the API below:

synthesizer = GaussianCopulaSynthesizer(
  metadata,
  numerical_distributions={
    'age': 'uniform',
    'weight': 'norm',
    'bmi': 'beta'
  }
)

The numerical_distributions parameter is only valid for the existing columns in my metadata. Currently, the synthesizer is allowing me to pass in something like:

numerical_distributions = {
  'totally_fake_column_name': 'beta'
}

Expected behavior

Check to make sure that the the dictionary keys in numerical_distributions actually correspond to column names (as specified by the metadata object). If they do not, then throw an error:

SynthesizerInputError: Invalid column names found in the numerical_distributions dictionary (<list>).
The column names you provide must be present in the metadata.

Eg. 
SynthesizerInputError: Invalid column names found in the numerical_distributions dictionary ('fake', 'fake2').
The column names you provide must be present in the metadata.