[Question/Design] Make Trial serializable with cloudpickle

Motivation

The design of study.optimize(callback, ..., n_jobs = ...) lends itself to parallel execution i.e. callback could be a remote procedure call.

However for this to work, Trial should be serializable whereas currently it is not (it holds some state which cannot be reconstructed across the wire).

Would this even be possible? What are the tradeoffs involved?

Description

This is just an exploratory question, but I imagine the change would entail some heavy refactoring.

Alternatives (optional)

No response

Additional context (optional)

No response