support dataset adapter

[xzrderek]

Note

Introduces adapter-driven dataset preparation for RFT and simplifies test resolution/selection.

• RFT dataset transformation: create_rft.py now detects a test’s dataset_adapter (via __ep_params__) and transforms raw JSONL into EvaluationRow JSONL before validation/upload (_extract_dataset_adapter, _maybe_transform_dataset_jsonl_via_adapter).
• Centralized module loading: New utils.load_module_from_file_path replaces ad‑hoc import logic in create_rft.py and upload.py.
• Test selection UX: Switches prompts to single-select; in non-interactive (--yes) mode proceeds only if exactly one test is found, otherwise errors with clearer messaging.
• Validation/flow tweaks: Dataset validation uses the adapted JSONL; data loaders are rejected for RFT dataset inference; minor resolver/logging cleanup.
• Tests: Adds coverage ensuring raw input datasets are adapted and validated before upload.

^{Written by Cursor Bugbot for commit 5a26ef7. This will update automatically on new commits. Configure here.}