[Enhancement] Activate Structured Outputs for AWS models

[iwr-redmond]

Motivation

The Amazon Web Services NLP adapter currently does not use structured outputs, which may result in errors during inference. AWS has recently introduced a Structured Outputs feature, which requires models to adhere to provided JSON schemas.

Solution Proposal

Modify the AWS NLP adapter to use Structured Outputs instead of standard prompts as described in

• This tutorial for commercial models
• Step 5 of this tutorial for custom models

Discussion

This FR is paired with #678, which would provide the same constrained sampling feature for user-provided open weights models.