fix(openai): warn about reasoningContent once per request - #5016

Human Overview

The "reasoningContent is not supported" warning was repeated for every earlier message on every request, logs filled up fast in long conversations; the check now runs once per request. LiteLLM also fixed. Once per request instead of once per model instance.

Description

OpenAIModel re-formats the whole history on every request and logged the reasoningContent warning once per reasoning message each time, so a reasoning-model run of N turns logged about N²/2 copies (552k in the reporter's benchmark). The check now runs once over the history, so each request logs at most one warning. LiteLLMModel shares this formatter and gets the same fix.

This warns once per request rather than once per model instance, as the issue suggested: the formatter is a classmethod with no instance to hold a "warned" flag, and the issue lists once per request as acceptable.

Resolves #4990

Related Issues

#4990, #1053 (how models handle reasoning content they didn't produce; this PR only changes how often the warning is logged)

Documentation PR

N/A — log-volume fix, no user-facing docs change.

Type of Change

Bug fix

Testing

Ran the issue's repro script: requests 1, 10 and 100 now each log 1 warning (100 in total over 100 requests, down from 5,050). Ran the OpenAI and LiteLLM model test suites and the full strands-py unit suite.

• I ran hatch run prepare

Additional Details

Checklist

• I have read the CONTRIBUTING document
• I have reviewed and understand every line of code in this PR, including any generated by AI tools, and I can explain why it works
• My change is focused and reasonably small; I have split unrelated work into separate PRs
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.