Unify sampling params for reject sampling

[npatki]

Problem Description

1. Our sampling methods assume sample (for all models) and GaussianCopula would never use reject sampling. This is incorrect. Due to constraints, any sampling method for any model might need reject sampling.
2. We are using batch_size and batch_size_per_try as interchangeable concepts when they are opposites: batch_size is intended to be small (to control output and save progress), but batch_size_per_try is intended to be large (for reject sampling)

Expected behavior

Have the same parameters for all sampling methods (sample, sample_conditions and sample_remaining_columns) for all single table models.

• batch_size which will be used in the intended way: Users should be able to split up large sampling tasks into smaller sizes
  • Requirement: batch_size <= total num_rows
  • Default: batch_size = total num rows
• max_tries_per_batch, which will be used when there is any kind of reject sampling involved
  • Default: 10

Logic

For every batch, for every try: Dynamically update the # of rows we try to sample to achieve the full batch_size.

TabularPreset

• Now that GaussianCopula has the same sampling params as other models, we need a special case for it. Make sure this works. (ref: Unable to sample conditionally in Tabular_Preset model #796)