fix: insert separator between Responses API phase commentary and final text

[ulugbekna]

Fixes #312163

Problem

When the Responses API streams multiple message output items — e.g. a commentary phase followed by a final_answer phase — the text from both items gets directly concatenated in the textAccumulator without any separator. This produces fused text like:

Responding directly in commentary as requested. My name is GitHub Copilot.My name is GitHub Copilot.

instead of properly separated paragraphs.

Fix

Track the output_index of the last response.output_text.delta event in OpenAIResponsesProcessor. When text arrives from a different output item, emit a \n\n paragraph separator first.

This only affects response.output_text.delta events — reasoning events (response.reasoning_summary_text.delta) use a completely separate code path and are unaffected.

Test

Added a test based on a real-world stream dump that replays the full SSE event sequence for a commentary → final_answer two-phase response and asserts the accumulated text is properly separated.

[Copilot]

Pull request overview

Fixes a Copilot Responses API streaming formatting issue where text deltas from separate output items (e.g., commentary then final_answer) were concatenated without separation, producing fused text in the accumulator.

Changes:

• Track the last response.output_text.delta output_index in OpenAIResponsesProcessor and insert a \n\n separator when the stream switches to a new output item.
• Add a unit test that replays a real-world SSE sequence for a commentary → final_answer two-phase response and asserts the accumulated text includes the separator.

Show a summary per file

File	Description
extensions/copilot/src/platform/endpoint/node/responsesApi.ts	Inserts a paragraph separator when output_text.delta events move to a different output_index to prevent phase text fusion.
extensions/copilot/src/platform/endpoint/node/test/responsesApi.spec.ts	Adds a regression test that replays a two-phase SSE stream and verifies the accumulator includes \n\n between phases.

Copilot's findings

• Files reviewed: 2/2 changed files
• Comments generated: 1

[Copilot]

This assertion depends on implementation details of how many progress callbacks fire near completion (using accumulatedTexts.length - 2). That’s brittle if processResponseFromChatEndpoint changes to emit fewer/more callbacks on response.completed. Consider asserting against the final accumulated text (e.g., last entry) or locating the last distinct/non-empty accumulated value before completion instead of indexing from the end.

Suggested change

	// The accumulated text must separate commentary and final_answer text
	const lastTextBeforeCompletion = accumulatedTexts[accumulatedTexts.length - 2];
	expect(lastTextBeforeCompletion).toBe(
	// The accumulated text must separate commentary and final_answer text.
	// Find the last meaningful accumulated value rather than depending on
	// how many callbacks fire during stream completion.
	const lastAccumulatedText = [...accumulatedTexts].reverse().find(text => text.length > 0);
	expect(lastAccumulatedText).toBe(