feat: preserve governed OpenAI tool messages in 1.4.0

Add faithful single-choice chat results, tool stream activity, error-frame retries, and isolated protocol caching.

Reject empty cached completions with the same value predicate as transport, add regression coverage, and adopt Apache-2.0.

tests/unit/test_openai_compat.py

@@ -90,6 +90,7 @@ async def _complete(
     reasoning_effort=None,
     first_token_event=None,
     content_event=None,
+    response_mode="text",
 ):
     return await transport.complete(
         messages=[{"role": "user", "content": "hi"}],
@@ -100,6 +101,7 @@ async def _complete(
         reasoning_effort=reasoning_effort,
         first_token_event=first_token_event,
         content_event=content_event,
+        response_mode="text",
     )
 
 
@@ -161,6 +163,8 @@ class TestSsePureFunctions:
         assert _sse_data_payload("data: [DONE]") == "[DONE]"
 
     async def test_iter_deltas_yields_and_flags_done(self):
+        from polygateway.transports._openai_completion import ChatCompletionAccumulator, ChatDelta
+
         async def lines():
             for raw in [
                 _chunk(content="he"),
@@ -173,16 +177,28 @@ class TestSsePureFunctions:
                     yield line
 
         sink = {}
-        out = [d async for d in _iter_sse_deltas(lines(), sink)]
-        assert out == [(True, "he"), (False, "think")]
+        out = [
+            d
+            async for d in _iter_sse_deltas(
+                lines(), sink, ChatCompletionAccumulator(response_mode="text")
+            )
+        ]
+        assert out == [ChatDelta("he", "", True), ChatDelta("", "think", True)]
         assert sink["done"] is True and sink["usage"] == _USAGE
 
     async def test_malformed_frame_raises_transient(self):
+        from polygateway.transports._openai_completion import ChatCompletionAccumulator
+
         async def lines():
             yield "data: {not-json"
 
         with pytest.raises(TransientError, match="malformed"):
-            _ = [d async for d in _iter_sse_deltas(lines(), {})]
+            _ = [
+                d
+                async for d in _iter_sse_deltas(
+                    lines(), {}, ChatCompletionAccumulator(response_mode="text")
+                )
+            ]
 
 
 class TestStreamHappyPath: