feat(datasets) Add VerticalEvenPartitioner

[adam-narozniak]

A new class that produces even or approximately even number of partitions (in terms of number of columns).

This class can't be a subclass of VerticalSizePartitioner due to the inability of accurate fraction (neither count) specification of sizes. Why?
e.g.

n_cols = 104
columns = [f"{i}" for i in range(n_cols)]
dataset = _create_dummy_dataset(columns, num_rows=50)
# If using the subclassing
partitioner = VerticalEvenPartitioner(num_partitions=100, shuffle=False)
partitioner.dataset = dataset
pd.Series(
    [partitioner.load_partition(i).num_columns for i in range(100)]
).value_counts()
# 1    99
# 5     1
# Name: count, dtype: int64

And here if using a new class

1    96
2     4
Name: count, dtype: int64

0     2
1     2
2     2
3     2
4     1
     ..
95    1
96    1
97    1

So why it is not possible to inherit?
an operation that creates either counts of fractions needs to happen at the init time (to pass the args to SizePartitinoer) However due to the late assignment of the dataset it's not possible to use count neither to provide a better way of assign fractions that [1/num_partitions] * num_partitions and the assign the remainder columns to the signed object.