refactor: use dataloader for table fields

[RogerHYang]

Note

Refactors GraphQL types and mutations/queries to use TableFieldsDataLoader-backed field resolvers and consistent Node(id, db_record) construction, while expanding annotation schema and related fields.

• GraphQL Schema/Types:
  • Expand Annotation interface and concrete types to include annotatorKind, metadata, source, identifier, user, and timing fields; align DocumentAnnotation, SpanAnnotation, TraceAnnotation, ExperimentRunAnnotation, and ProjectSessionAnnotation accordingly (e.g., traceId: ID).
  • Migrate many types (Project, Trace, Span, Dataset*, Prompt*, User*, etc.) to lazy, dataloader-backed field resolvers; standardize construction to Node(id, db_record).
• DataLoaders/Context:
  • Add numerous TableFieldsDataLoader entries (e.g., for datasets, splits, versions, prompts, labels, users, api keys, annotations, experiments, traces, sessions) and expose them through Context.DataLoaders.
• Mutations/Queries:
  • Update mutations (prompts, prompt labels/tags, datasets/splits/labels, span/trace/session annotations, users, API keys) to return nodes via id and optional db_record instead of bespoke converters.
  • Refactor prompt creation: move prompt version normalization into ChatPromptVersionInput.to_orm_prompt_version and simplify prompt mutations.
  • Adjust queries and node resolver to construct nodes with id and rely on dataloaders for fields; remove legacy to_gql helpers.
• Runtime/Subscriptions:
  • Update chat completion (mutation/subscription) payloads to use Span(id, db_record) and enqueue events as before.

^{Written by Cursor Bugbot for commit 59da6ad. This will update automatically on new commits. Configure here.}

[axiomofjoy]

It might be worth making the interface classes into abstract base classes with abstract methods to ensure the sub-classes implement the required fields.

[axiomofjoy]

Suggested change

	@strawberry.field
	async def color(
	self,
	info: Info[Context, None],
	) -> str:
	if self.db_record:
	val = self.db_record.color or "#ffffff"
	else:
	val = await info.context.data_loaders.dataset_split_fields.load(
	(self.id, models.DatasetSplit.color),
	)
	val = val or "#ffffff"
	return val
	@strawberry.field
	async def color(
	self,
	info: Info[Context, None],
	) -> str:
	if self.db_record:
	val = self.db_record.color
	else:
	val = await info.context.data_loaders.dataset_split_fields.load(
	(self.id, models.DatasetSplit.color),
	)
	return val

Required type

[axiomofjoy]

Suggested change

	self, info: Info[Context, None]
	self

[axiomofjoy]

The Annotation interface has more in it than it used to. Is that on purpose?