I would love if FeatureReader could read data from an attribute, not just from disk file (and the same way, FeatureWriter could write the contents to an attribute instead of a disk)

Eg. binary data files (image, PDF, Excel, etc) are data URI encoded in JSON from a web service. It is straightforward to extract and decode back to binary to an attribute. However, this binary must be written to disk and read back by a FeatureReader. Writing and reading a disk is always slower than memory and it would be one less dependency (eg. folder and filename) to take care of if the FeatureReader could read from an attribute.