Add multimodal outputs to js_repl polling

[fjord-oai]

Summary

Adds multimodal output support to js_repl polling results.

This is a follow-up to the polling-core change: completed js_repl_poll calls can now return structured content items, so nested tools like view_image work cleanly in polling mode instead of being flattened to text-only JSON.

What Changed

Poll output formatting

• Updated js_repl_poll result formatting so completed polls can return FunctionCallOutputBody::ContentItems
• When multimodal output is present, the poll result now emits:
  • a first input_text item containing the JSON status payload
  • followed by any nested multimodal content items (for example input_image)

Polling runtime

• Extended polling exec state to retain nested tool-call content items until completion
• Completed polling execs now surface those content items through JsExecPollResult
• Replayable completed polls preserve the same multimodal output shape on subsequent polls

Docs / guidance

• Updated tool spec text, project-doc guidance, and docs/js_repl.md to document that completed polls may include nested multimodal output after the JSON status item

git stack

• ⏳ 1 Add feature-gated js_repl polling flow #10673
• 👉 2 Add multimodal outputs to js_repl polling #13081