Test loading S3 parquet folder in dataset format

[mzwiessele]

Test that a parquet dataset saved in dataset format can be loaded and accessed using get_historical_features.

Steps:

• upload a parquet dataset to the test s3 folder
• point a source to the parquet dataset s3 folder (not the file)
• load the source features using get_historical_features

Ideally, there'd be a test for this running, but you'd be blocked on making a file within the Feast integration tests S3 bucket. Mind making an issue for that separately? @mzwiessele

Originally posted by @adchia in #3217 (comment)

[felixwang9817]

I think there are probably two ways to do this:

• there is an S3FileDataSourceCreator in sdk/python/tests/integration/universal/feature_repos/universal/data_sources/file.py, which uses minio - it hasn't been used in a while, but it could be updated and used
• instead of using minio, we could also just create a dataset and upload data directly to an S3 bucket - we could just use the feast-integration-tests S3 bucket, which is used in universal/data_sources/redshift.py

@mzwiessele if you want to take this on, I'd be happy to help you and review any PRs!