Add vendor-neutral IR core library

[vaayne]

Summary

• implement a vendor-agnostic IR data model, builder helpers, and metadata utilities for PolyLLM
• add validation, normalization, and streaming aggregation support to convert events into responses
• document the new IR package structure and usage in a dedicated README

Testing

• go test ./ir

---

https://chatgpt.com/codex/tasks/task_e_68de5f9d43508322b6e3042cabd47227

[vaayne]

@codex You should implement @plan.md using TDD ， check current code and using TDD to continue

[chatgpt-codex-connector]

Summary

• Enforced ValidateRequest to reject unsupported message roles, ensuring IR requests adhere to the defined role set before processing. ir/validate.goL157-L176
• Added focused unit tests that cover both failure and success paths for message-role validation to guard against regressions. ir/validate_test.goL1-L45

Testing

• ✅ timeout 5 go test ./ir

View task →

[vaayne]

@codex review implement base on @plan.md， MUST make sure all implementations follow the plan and the implementation is using TDD pattern

[chatgpt-codex-connector]

Summary

• Restricted ValidateRequest to accept only the planned user, assistant, and tool roles, returning a field error for unsupported values. ir/validate.goL157-L174
• Added unit tests covering both rejection of unsupported roles and acceptance of allowed roles to anchor the validation behavior. ir/validate_test.goL5-L37

Testing

• ✅ go test ./ir

View task →

[vaayne]

@claude can you check @plan.md and implement this