Add ability to reset random sampling

[amontanez24]

Problem Description

As a user, it would be nice if I could get the same sampled results as a previous sample call. It would also be nice to control when since sometimes I want new sampled data. On top of this, the consistency should be for all columns.

Acceptance criteria

• We want to change the way we handle randomness to make the above request possible. This means doing the following steps:
  • Remove the randomize_samples parameter from all sample calls
  • Set the seed for the underlying model on the initial call.
  • Add a method called reset_sampling that resets the seed for the model back to the original state, as well as resets that random state of the HyperTransformer.
  • We should add the same method to MultiTableSynthesizers as well. This would just loop through each SingleTableSynthesizer and call the method.

Expected behavior

from sdv.single_table import GaussianCopulaSynthesizer

synthesizer = GaussianCopulaSynthesizer(metadata)
synthesizer.fit(data)
synthetic_data_1 = synthesizer.sample(10)

synthesizer.reset_sampling()
synthetic_data_2 = synthesizer.sample(10)

synthetic_data_2 should be the same as synthetic_data_1

Additional context

• We may need to add some logic to track the state of randomization for the underlying model. Currently, we have a method to set the random state on them. We should set the random state in the beginning and then let the model continue to use that state until reset_sampling is called