fix: use external openai NotGiven/Omit types for compatibility

[roh26it]

Problem

When using Portkey with libraries like pydantic-ai that pass NOT_GIVEN values from the standard openai package, users encounter:

TypeError: Object of type NotGiven is not JSON serializable

This happens because Portkey's vendored OpenAI SDK defines its own NotGiven class, which is a different class from the standard openai's NotGiven. When pydantic-ai passes values using the standard openai's NOT_GIVEN, the vendored code doesn't recognize them as NotGiven instances.

Solution

Instead of defining duplicate NotGiven and Omit classes in the vendored code, we now:

1. Try to import from the external openai package if it's installed
2. Fall back to local definitions only if openai is not installed

This ensures that when libraries like pydantic-ai pass NOT_GIVEN values from the standard openai package, Portkey's vendored SDK recognizes them because they're the exact same class.

Testing

import openai._types as external_types
import portkey_ai._vendor.openai._types as vendored_types

# Now they're the same class
assert external_types.NotGiven is vendored_types.NotGiven  # ✓
assert external_types.NOT_GIVEN is vendored_types.NOT_GIVEN  # ✓

# isinstance checks work correctly
from portkey_ai._vendor.openai._utils import is_given, strip_not_given
assert is_given(external_types.NOT_GIVEN) == False  # ✓
assert strip_not_given({'a': 1, 'b': external_types.NOT_GIVEN}) == {'a': 1}  # ✓

[Copilot]

Pull request overview

This PR resolves compatibility issues between Portkey's vendored OpenAI SDK and external libraries (like pydantic-ai) that use the standard openai package's NOT_GIVEN sentinel values. The fix prioritizes importing NotGiven, Omit, and NOT_GIVEN from the external openai package when available, falling back to local definitions only when openai is not installed.

Key changes:

• Added try/except block to import NotGiven, Omit, and NOT_GIVEN from external openai package
• Ensured compatibility by using the same class instances across packages
• Maintained backward compatibility with standalone usage when openai is not installed

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Creating a new instance of Omit() in line 127 may not maintain compatibility if the external openai package uses a singleton pattern for omit. Consider checking if the external package exports omit as a module-level constant and using that instead of creating a new instance.

Suggested change

	User code shouldn't need to use not_given directly.
	# External openai only exports NOT_GIVEN, not the lowercase not_given
	not_given = NOT_GIVEN
	# External openai doesn't export lowercase omit, create from class
	omit = Omit()
	except ImportError:
	# Fall back to defining our own (for standalone use without openai installed)
	# Try to import the singleton 'omit' if available
	try:
	from openai._types import omit as _external_omit
	omit = _external_omit
	except ImportError:
	# If not exported, create from class
	omit = Omit()
	# External openai only exports NOT_GIVEN, not the lowercase not_given
	not_given = NOT_GIVEN