fix(client): drop v1-incompat discriminator on InputItem (b48) (#814)

InputItem's __root__ union uses Field(discriminator="type"), but one of
its submodels (Item) is itself a RootModel union whose nested submodels
don't satisfy pydantic v1 compat's strict discriminator requirements.
Under pydantic v2 with v1 compat, class definition fails at import:

  pydantic.v1.errors.ConfigError: Field 'type' is not the same for all
  submodels of 'Item'

This is the same fix Brandon applied in 1d50898 (feat: support openai 2
in python client), which was wiped out by a subsequent codegen regen.

Dropping the discriminator on InputItem makes it a regular Union (try-
each validation). Valid data still deserializes into the correct submodel
because each has a Literal[...] type field — only the matching submodel
validates. JSON input/output is byte-identical.

Minimal patch. Other 9 discriminator="type" sites in gen/openai.py are
left intact; they don't recursively reference RootModel unions so they
don't hit the v1-compat failure.

Local verification:
- Before: `from llmengine import Completion, Model` raises ConfigError
- After: imports cleanly; ChatCompletionV2Request/Response work; all 9
  Item submodels import; InputItem/Item/ItemResource/OutputItem1 import;
  InputMessage instantiation + validation roundtrips

Follow-ups:
- Fix OpenAPI schema or datamodel-code-generator config so next regen
  doesn't re-introduce the bug
- Add import smoke test to llm-engine CI under pydantic 2 + v1 compat

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: scale-ballen

clients/python/llmengine/data_types/gen/openai.py

@@ -11516,7 +11516,7 @@ class EvalRunList(BaseModel):
 
 class InputItem(BaseModel):
     __root__: Annotated[
-        Union[EasyInputMessage, Item, ItemReferenceParam], Field(discriminator="type")
+        Union[EasyInputMessage, Item, ItemReferenceParam], Field()
     ]
 
 

clients/python/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "scale-llm-engine"
-version = "0.0.0.beta47"
+version = "0.0.0.beta48"
 description = "Scale LLM Engine Python client"
 license = "Apache-2.0"
 authors = ["Phil Chen <phil.chen@scale.com>"]