Add Dataset Split

[kapkic]

Even though we don't have Classification or any other task that'd require splits, we will need to implement it in the future.
Goal is to define test-train, or any kind of dataset split support.

[Shu-Wan]

here are some scenarios to consider

task 1: single graph causal discovery (discovery.static or discovery.temporal)

• no split required

task 2: multiple graphs causal discovery

• split based on files (datasets, graphs)

task 3: (generic, supervised learning) classification & regression

• single tabular file: split on row index
• multiple file: split based on files (datasets, file dictionary)

task 4: (generic, supervised learning) spatial / temporal classification & regression

• split based on the time/location index

so we need to support split based on

• row index
• some label
• file index

anything else?