feat: change Prompt integer variants from u16 to u32 for future compatability

[paulhendricks]

Summary

Updates the Prompt enum to use u32 instead of u16 for integer-based prompt representations. The change affects both Prompt::IntegerArray and Prompt::ArrayOfIntegerArray variants, as well as associated impl_from_for_* macro implementations.

Motivation

The original use of u16 may have been based on the assumption that token ID values would not exceed the vocabulary size of models like GPT-2 or GPT-3 (i.e. 50256 tokens).

While that assumption held previously, newer models frequently exceed this range. Allowing u32 values would enable inference libraries to use u32 token IDs and larger vocabulary sizes without needing to truncate while batching.

I fully recognize this is a breaking change to the public API:

• Users of Prompt::IntegerArray or Prompt::ArrayOfIntegerArray with u16 inputs will encounter compile-time errors
• Pattern matches and constructors relying on the u16 variant types must be updated to use u32

However, I think that this is in line with the scope and mission of async-openai as the OpenAI OpenAPI schema does not specify the maximum token ID value.

And while users of Prompt::IntegerArray or Prompt::ArrayOfIntegerArray with u16 inputs will need to update, conversion of u16 to u32 is lossless and won't cause any runtime issues. And the benefits of future proofing and greater compatibility outweigh prioritizing smaller vocabulary sizes from the previous GPT-2/GPT-3 era.

Thank you for your consideration!

[ishandhanani]

Hi @64bit - can we get a review on this? Would love to use this for https://github.com/ai-dynamo/dynamo

[64bit]

Thank you for detailed context, appreciate it! @paulhendricks

@ishandhanani glad you find it useful!