Function to convert ragged arrays into (Python) _lists_ of Tensors?

[jpivarski]

Description of new feature

Similar to #3256 and #3257, another function that could help ML users is one that turns a ragged array into a simple Python list of tensors. Although the aggr functions in PyTorch-Geometric can use ak.parents_index from #3256 to do a segmented aggregation in one Python call, tools like PyTorch Cluster expect an explicit Python loop over each variable-sized list (e.g. a Python loop over events, calling a vectorized function to process each collection of particles in that event: they're "batched by event.")

Perhaps ak.Array → list[torch.Tensor]? Or → DataLoader (batched by lists in the ragged array)?

Cc: @livaage, @GageDeZoort, @maxymnaumchyk, @ianna