Extend sample handling cases in linear_pool

[lshandross]

The initial implementation of sample handling in linear_pool assumes that every unique combination of task IDs for the same model has the same number of provided component samples and target component samples. This assumption may not always hold true, so we would like to instead allow for varying numbers of samples for both provided and target component samples, which will require mapping over unique task ID combinations instead of models

Further, the initial implementation uses stratified sampling by models, but we may be interested in being able to specify one of various methods in an argument to the function, as raised in Issue #27