Support input and output transformation in current inference serving script

Users sometimes need to change the "tabular_serve.py" to filter the data and also return the output in a particular format. The same logic should be allowed for all modality using customizable input and output transformer

Start by reading tabular_serve.py and tracing how inference inputs and outputs are currently handled. Determine how customizable transformers could filter inputs and format outputs consistently across all modalities; done means the serving script supports both transformations for each modality.

Written by the indexing model from the issue text.