Dataset / agree on what `load_by_name` does

In the new dataset, several people have already been bugged by the facts that

• An experiment and a run can have any non-unique name
• load_experiment_by_name expects a unique name

(see e.g. #991)

I think we have an API in conflict with itself, which I suggest that we change. There are two obvious ways to go (with the first way branching into two ways). Either

• An experiment and a run are forced to have a unique name (in one of two ways)
  • This happens by raising an error when a name i reused
  • This happens by automatically changing the name to a unique one (prepending a number or so)
• load_experiment_by_name expects a unique name

or

• An experiment and a run can have any non-unique name
• load_experiment_by_name loads all experiments/runs that match the given name

So, which one do you prefer? 1a, 1b, or 2?

@QCoDeS/core @jana-d @majacassidy

Start by reviewing the dataset API entry point for load_experiment_by_name and the related context in #991; the issue names no file or test. This issue is not ready for implementation until the project chooses among 1a, 1b, and 2 and defines the expected behavior for duplicate names.