Feature Request: Support sparse to dense with max_length input

[chenghuige]

Using tensorflow 2.3, some times I found I need tfrecords with tf.io.VarlenFeature, but when using tpu or for some other reasons you might need to pad it to fixed max_length like 50, 200(fixed) not just using padded_batch or tf.sparse.to_dense to padd it to the max length in the batch(dynamic).

It is ok if you pre padding to max_length when generating tfrecords but that is wasting space and also reduce flexbilty.

I'd like padded_batch(x, max_length), tf.sparse.to_dense(x, max_length) or just set tf.io.VarlenFeature(max_length).

Will this feature be possible ? Thanks!