[FEA] Smart Tensor Creation with File Reading APIs

[tylera-nvidia]

Is your feature request related to a problem? Please describe.
Currently all of the FILE_IO APIs require the user to size a tensor completely (RANK and dim sizes) before handing the tensor to the read_* APIs to ingest a given file. This requires either:

Describe the solution you'd like
Update the read_* APIs to support either creation of a tensor on read, or dynamic resizing of a tensor with defined RANK.

Describe alternatives you've considered
Currently, the read functions require either:

• prior knowledge of statically sized files
• additional file parsing/wrangling from the user to dynamically size their tensor.

both scenarios are either limiting/burdensome