Add in-loop gen tasks

[davidheineman]

This PR adds the BPB over the correct continuation in GSM, Minerva, MBPP and HumanEval.

The BPB task configs use the setup implemented in https://github.com/allenai/oe-eval-internal/pull/374, and the implementation is the same as allenai/OLMo#777

Evaluation Setup

The "gold continuation" appears as follows:

Question: Janet’s ducks lay 16 eggs per day. She eats three for breakfast every morning and bakes muffins for her friends every day with four. She sells the remainder at the farmers' market daily for $2 per fresh duck egg. How much in dollars does she make every day at the farmers' market?
Answer: Janet sells 16 - 3 - 4 = 9 duck eggs a day. She makes 9 * 2 = $18 every day at the farmer’s market. So the answer is 18.

Sanity check

I tested this with the in-loop trainer:

torchrun --nproc-per-node=4 src/scripts/train/OLMo2-1B.py train test_run .

[undfined]

lgtm

Nit: Do the json configs need to be unformatted? If not then a lot easier to read/scan down the road.

I guess one other nit: We're kind of mixing and matching prefixing these with codex_ here and in oe-eval-internal. Would be nice to be consistent.

[davidheineman]

Thanks for taking a look! Updated mbpp -> codex_mbpp.

Agree on your first nit: but I won't format the .json in this PR, since the other .json files are not formatted.