LLM output uniformizer for all providers

[mschoenb97IL]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Right now some of the message chunks from streaming output are in a different format for different LLM providers. For example, Anthropic gives a list of chunks while openai gives a string. Furthermore, llama models from ChatBedrockConverse have AIMessage content fields as lists rather than strings. Can these outputs either be made uniform, or can you expose a utility to uniformize them?

Motivation

I want to be able to easily switch between providers without downstream processes breaking

Proposal (If applicable)

No response

[baskaryan]

I think adding a utility makes the most sense, at least to start. Will open PR shortly, would love feedback!

[baskaryan]

#25832