[TensorFlowLite] Add new data types in the TFLite format schema

[mciprian13]

Add new data types in the TFLite format schema that could be leveraged by custom accelerators, such as:

• Custom floating point format, having configurable number of bits for exponent and mantissa (something like TensorType_FLOAT8 using custom format options)

This is relevant only for the TFLite format (schema) used for model representation and not necessary for the TFLite tools or runtime.
This is relevant when industry has proprietary tools that rely on the TFLite format but not on the TFLite tools or runtime (interpreter),

[pkgoogle]

Seems like a reasonable request, @miaout17, can you please take a look? Thanks.