Optimize OpenID configuration fetching with LRU cache

Summary

This PR optimizes the get_discovery_doc function to reduce latency and redundant network requests when creating multiple AuthClient instances. The changes include:

• Added @lru_cache() decorator to cache discovery document responses per environment
• Removed the session parameter dependency from get_discovery_doc
• Always use requests.get() directly since authentication is not needed for OpenID configuration endpoints
• Updated the call site in AuthClient.__init__ to remove the session parameter
• Fixed the test suite by clearing cache before mocking bad responses

This optimization will significantly improve performance for applications that create multiple AuthClient instances, as the discovery document will only be fetched once per environment and then served from cache.

Review & Testing Checklist for Human

• Verify discovery endpoints don't require session auth: Confirm that .well-known/openid_configuration/ endpoints work correctly without session-specific configurations (cookies, proxies, auth headers)
• Test caching behavior: Create multiple AuthClient instances and verify they share the cached discovery document correctly
• Test long-running applications: Verify that cached discovery documents don't cause issues if the actual endpoints change (though this should be rare)
• Test custom environment URLs: Ensure caching works correctly with both predefined environments ('sandbox', 'production') and custom URLs

Notes

• The discovery document cache persists for the lifetime of the process - this is intentional for performance but worth being aware of
• All existing tests pass, including the modified test that now clears cache before mocking
• Link to Devin run: https://app.devin.ai/sessions/711c681a2e0c42d880347d75b025c6dc
• Requested by: @abhay-codeium