Remove evaluator state

[snarayan21]

What does this PR do?

We don't use eval state at all -- currently we simply set the eval dataloader state only if the dataset is a StreamingDataset, and we also simply set the current sample to 0 (meaning that we start from the beginning). This is not needed at all and is causing some errors when resuming training on various numbers of GPUs. Since we iterate through the entire eval loader every single time we do eval, there is no notion of state that has to be preserved for eval datasets. This PR removes that functionality completely.

Specifically, this caused errors with resuming on various numbers of GPUs and setting device_eval_batch_size. When going from 24 to 32 gpus, with both runs specifying device_eval_batch_size of 2, Streaming would try to resume using a global eval batch size of 64 samples, but the state dict would also show that 24 devices had been used in the initial run. To deterministically resume, Streaming attempts to partition the current global batch size over the initial number of devices, and 64 samples are not divisible by 24 devices, throwing an error.

Test runs have succeeded:
1b-dense-fsdp-shardgradop-lion8b-fullckpt-resume-s44ELm: resume from old checkpoint correctly
1b-dense-fsdp-shardgradop-lion8b-fullckpt-start-HDVoIC: save new checkpoint with PR
1b-dense-fsdp-shardgradop-lion8b-fullckpt-resume-VPH2X6: resume from new checkpoint correctly

What issue(s) does this change relate to?

Before submitting

• Have you read the contributor guidelines?
• Is this change a documentation change or typo fix? If so, skip the rest of this checklist.
• Was this change discussed/approved in a GitHub issue first? It is much more likely to be merged if so.
• Did you update any related docs and document your change?
• Did you update any related tests and add any new tests related to your change? (see testing)
• Did you run the tests locally to make sure they pass?
• Did you run pre-commit on your change? (see the pre-commit section of prerequisites)

[mvpatel2000]

Can you describe the errors this caused?

[snarayan21]

@mvpatel2000 updated PR description. I'm not going to merge this until Kushal can resolve the issue with this branch

[snarayan21]

Test runs succeeded (see PR desc), merging