Allow openai 3.x in llama-index-llms-openai

The `openai<3` cap was added in 0.7.0 (March 2026), five months before
openai 3.0.0 shipped, so it was a routine next-major cap rather than a
reaction to a known incompatibility. openai 3.x has been out since
August 2026 and the package still works against it.

Testing against openai 3.7.0 with the cap removed:

* The full package suite passes on both openai 2.54.0 and 3.7.0,
  including the API-backed tests, which cover streaming, async
  streaming, structured prediction, built-in tools and file upload.
* All type and client import paths the package uses still resolve, and
  no parameter it passes has been removed.
* openai 3.0.0's one documented breaking change is the move to HTTPX2.
  The package never constructs a transport itself, it only forwards an
  optional `http_client`, and openai 3 accepts both client families.
  Requests, including streaming, succeed either way.

One test needed updating. `ResponseOutputTextAnnotationAddedEvent.annotation`
was typed as a bare `object` in openai 2 and is a discriminated union in
openai 3, so the fixture's placeholder payload no longer validates. It
now uses a real annotation and normalises the stored value, which works
on both majors. No library code required changes.

Claude-Session: https://claude.ai/code/session_01YDFeREeB9bZQ3aeuKfJses

Author: BenA-SA

llama-index-integrations/llms/llama-index-llms-openai/pyproject.toml

@@ -27,13 +27,13 @@ dev = [
 
 [project]
 name = "llama-index-llms-openai"
-version = "0.8.0"
+version = "0.9.0"
 description = "llama-index llms openai integration"
 authors = [{name = "llama-index"}]
 requires-python = ">=3.10,<4.0"
 readme = "README.md"
 license = "MIT"
-dependencies = ["openai>=1.108.1,<3", "llama-index-core>=0.14.5,<0.15"]
+dependencies = ["openai>=1.108.1", "llama-index-core>=0.14.5,<0.15"]
 
 [tool.codespell]
 check-filenames = true

llama-index-integrations/llms/llama-index-llms-openai/tests/test_openai_responses.py

@@ -292,7 +292,13 @@ def test_process_response_event_with_text_annotation():
         content_index=0,
         annotation_index=0,
         type="response.output_text.annotation.added",
-        annotation={"type": "test_annotation", "value": 42},
+        annotation={
+            "type": "url_citation",
+            "start_index": 0,
+            "end_index": 5,
+            "title": "Example",
+            "url": "https://example.com",
+        },
         sequence_number=1,
     )
 
@@ -307,9 +313,13 @@ def test_process_response_event_with_text_annotation():
     # The annotation should be added to additional_kwargs["annotations"]
     _, _, updated_additional_kwargs, _, _, _ = result
     assert "annotations" in updated_additional_kwargs
-    assert updated_additional_kwargs["annotations"] == [
-        {"type": "test_annotation", "value": 42}
-    ]
+    assert len(updated_additional_kwargs["annotations"]) == 1
+    stored = updated_additional_kwargs["annotations"][0]
+    # openai<3 types this field as bare ``object`` and leaves the dict alone;
+    # openai>=3 parses it into a discriminated annotation model.
+    stored = stored if isinstance(stored, dict) else stored.model_dump()
+    assert stored["type"] == "url_citation"
+    assert stored["url"] == "https://example.com"
 
 
 def test_get_tool_calls_from_response():