[EAGLE-3698] - model upload handles multiple batch

[phatvo9]

Why

For now model only predicts one by one input even sending a batch with size >1.

How

• get_predictions() method in inference.py will take a list of inputs instead of single input.
• Update examples with batch input.
• Update doc

Other updates:

• insert triton decorator function when initializing model repository, so user won't forget to do this.
• enable infer param description from_kwargs.

Note:

Models generated by a lower version will not function on this version.

[HarmitMinhas96]

This PR is currently very large. Can it be broken down into more manageable chunks for review?
E.g.:

1. Add multiple batches handling in one PR (maybe two if it can be logicially split)
2. Add new model type example text-embedder
3. Add new model type example multimodal-embedder
4. Add vllm example

Or the model type and vllm examples can be added first if you prefer

[phatvo9]

This PR is currently very large. Can it be broken down into more manageable chunks for review? E.g.:

1. Add multiple batches handling in one PR (maybe two if it can be logicially split)
2. Add new model type example text-embedder
3. Add new model type example multimodal-embedder
4. Add vllm example

Or the model type and vllm examples can be added first if you prefer

Broke down into #236 : update code for batching and update old examples and #237 added text-embedder and multimodal-embedder examples
Since vllm example is merged #217, so I added it together with old examples.

[phatvo9]

Closed it since #236 and #237 merged