feat(llmobs): add SyncExperiment.rerun_evaluators() and make task/dataset optional

[mehulsonowal]

Summary

• Add SyncExperiment.rerun_evaluators() to re-score stored task outputs with updated evaluators without re-running the task function
• Make task and dataset optional in LLMObs.experiment() / Experiment.__init__ / SyncExperiment.__init__ to support the pull() + rerun_evaluators() workflow
• Guard Experiment.run() to raise ValueError when called without task and dataset
• Add missing_task_strategy parameter ("raise" / "skip" / "retry") to control error-row handling during re-runs
• Add tests covering: rerun skips task execution, error strategies, result object replacement, init without task/dataset, run() guard, and rerun without task/dataset

Behavior

# Standard workflow (unchanged)
exp = LLMObs.experiment("my-exp", task, dataset, evaluators=[my_eval], project_name="p")
result = exp.run()

# Re-run evaluators only — no task re-execution
new_result = exp.rerun_evaluators()

# Future: pull prior results from backend then re-evaluate
exp = LLMObs.experiment("my-exp", evaluators=[updated_eval], project_name="p")
exp.pull()  # sets exp.result (pending backend endpoint)
new_result = exp.rerun_evaluators()

Test plan

• test_experiment_rerun_evaluators_skips_task — verifies _process_record is not called again
• test_experiment_rerun_raises_on_task_errors — default "raise" strategy
• test_experiment_rerun_skip_strategy — "skip" skips error rows
• test_experiment_rerun_invalid_strategy — invalid strategy raises ValueError
• test_experiment_rerun_no_prior_result — guards against calling before run()
• test_experiment_rerun_preserves_experiment_id — no new experiment created
• test_experiment_init_without_task_and_dataset — no error at init (passes locally)
• test_experiment_run_without_task_raises — run() guard (passes locally)
• test_experiment_rerun_without_task_succeeds — rerun works with pre-loaded result

🤖 Generated with Claude Code

[cit-pr-commenter-54b7da]

Codeowners resolved as

releasenotes/notes/llmobs-experiment-rerun-evaluators-b3c4d5e6f7a8b9c0.yaml  @DataDog/apm-python