Since version 0.16.5, th following code returns a null worksheet:
XLSX.utils.json_to_sheet([], {header: columns})

Is it now working as it is wanted? If so, how could we create an empty worksheet? In my case it is expected that sometimes the data is an empty array. And an empty worksheet only with headers is expected.

Pressing ENTER key has no effect on dialog for adding a new column containing entity identifers (from the Reconcile menu)

To Reproduce

Steps to reproduce the behavior:

1. Reconcile a few cells in a column (already done, see attached OpenRefine project to import)
2. On Species column, choose Reconcile -> Add entity identifiers column
3. Type in a new column name into the input box and p

https://www.audiomack.com/data-api/docs

See starter code in #681

This feature would be needed for at minimum the CelebA dataset.

Nested structures could be stored on disk and documented in schema in a "flat" way, but have nesting in as_dataset and in the FeatureDict (for data gen). Whenever applicable, nesting / un-nesting should be consistent with [`tf.contrib.framework.nest.

When using e.g. AvroSortedBucketIO, if a file pattern is passed as part of the input path, scio should output a helpful error message. Current behavior in 0.9.3 leads to a NullPointerException.

val myEntityPaths: List[String] = List(
  "gs://bucket/dir1/*.avro",
  "gs://bucket/dir2/*.avro",
)
sc.sortMergeCoGroup(
  classOf[String],
  AvroSortedBucketIO
    .read(new TupleTag[MyE