Refactor FixedSampleInfoDataset to Support Multi-Input Models

[reuvenperetz]

Pull Request Description:

The previous implementation of FixedSampleInfoDataset caused issues when working with models that expected multiple inputs with different shapes. Specifically, during inference it would raise a ValueError indicating a mismatch between the expected and received number of input tensors.
Instead of directly converting samples to tensors, we now separate the input tensors by their position in the input tuple, create a dictionary mapping each tensor position to its corresponding data, use tf.data.Dataset.from_tensor_slices with the dictionary structure and map the dictionary outputs back to the expected tuple format.
This preserves the tensor grouping and shapes while allowing different shapes for different input layers.

Checklist before requesting a review:

• I set the appropriate labels on the pull request.
• I have added/updated the release note draft (if necessary).
• I have updated the documentation to reflect my changes (if necessary).
• All function and files are well documented.
• All function and classes have type hints.
• There is a licenses in all file.
• The function and variable names are informative.
• I have checked for code duplications.
• I have added new unittest (if necessary).

[ofirgo]

Please wait for this #1321 PR to be merged before merging this one