Given a sequence of rows, partition into an undefined number of partitions of at most N rows but keep the header row as the first for all sequences.

Given a sequence of string[] rows, parse into columnar data. See csv->columns. This method is useful if you have another way of generating sequences of string[] row data.

Sample at most N rows selected randomly from the row sequence. If sequence is shorter than length N will return less than N rows. Uses naive reservoir sampling: https://en.wikipedia.org/wiki/Reservoir_sampling