add custom_ds_n CLI arg

[reaganjlee]

Allows --custom_ds_n [custom_name] to shorten long naming from multiple datasets

Unmodified, the output path to eliciting with datasets super_glue boolq and christykoh/boolq_pt
elk-reporters/bigscience/bloomz-7b1-mt/christykoh/boolq_pt, super_glue boolq/quizzical-mccarthy

To cd to this reporter requires, elk-reporters/bigscience/bloomz-7b1-mt/christykoh/"boolq_pt, super_glue boolq"/quizzical-mccarthy where path is divided with arbitrary number of / in the datasets as opposed to where datasets are actually divided

You can use this keyword when running elk elicit/eval without typing the full path to group the experiments by datasets, such as in the example above, --custom_ds_n multiboolq to get elk-reporters/bigscience/bloomz-7b1-mt/multiboolq/quizzical-mccarthy

[lauritowal]

random unimportant thing: because we now require Python >= 3.10, we can replace Union[X, Y] with X | Y and Optional[X] with X | None in type annotations

[lauritowal]

I think we would need to do the same for models too.
However, I feel like, what we want instead maybe, is to be able to set a custom path for the whole root ... or at least what follows after elk_reporter_dirt()

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[norabelrose]

hmm it's not really clear to me that we need this in addition to the custom out_dir