Pass through case version tags from R json

[alistaire47]

Closes #107. When it exists, pulls the value for case_version out of the tags of the JSON generated by {arrowbench} when running R benchmarks, and appends it to the tags sent on to conbench.

Notably this PR does not replace all the Python-generated tags with the R-generated ones, because they do not correspond exactly, so doing so would break histories. There are reasons they differ, e.g. {arrowbench} does not also append name as a tag (maybe it should for history's sake? but it's duplicative); params vary in cases like csv-read where Python and R do not have the same params and functionality. Long-term, we'll need to fix the situation, but efforts now to align {benchmarks} and {arrowbench} will become historical artifacts as orchestration gets moved out of {benchmarks}, so I think it's more worthwhile now to figure out what we want our data to look like, and then break all our histories at that point. Open to other ideas!

[jonkeane]

This looks good. How hard would it be to add a test that covers this circumstance + that it's going what we expect?

[alistaire47]

How hard would it be to add a test that covers this circumstance + that it's going what we expect?

A little hard, because arrowbench doesn't have an exported benchmark with case versioning (they're constructed directly during testing). _example_benchmarks.py (used for testing) currently references placebo, so we could add a versioned copy of that, but I don't see a way to do this neatly without a. a PR to arrowbench, b. including some R code in {benchmarks}, or c. some serious mocking.

If we want to go with (b) or (c) that's fine, but we should probably do it as part of a larger revamp in our unit testing where we test the core structures of {benchmarks} more directly than we do currently.

[jonkeane]

nods ok, would you mind making a TODO with some of those options for testing so we remember to come back to that when we do that larger revamp? Then I'm fine to merge this as is