Add training scripts for all-future and next-step supervision with DeepHealth

- Implement `train_all_future.py` for training with query-conditioned all-future supervision.
- Implement `train_next_step.py` for training with next-token/next-time-point supervision.
- Introduce `train_util.py` for shared utility functions including logging, dataset splitting, and model checkpointing.
- Enhance argument parsing for both training scripts to accommodate new parameters.
- Update loss functions and model configurations to support the new training paradigms.

evaluate_auc.py

@@ -18,7 +18,7 @@ Efficiency notes:
     avoiding repeated pickling of arrays for every disease.
 
 Run from the DeepHealth code directory containing dataset.py, models.py,
-readouts.py, and train.py-compatible checkpoints/configs.
+readouts.py, and train_config.json-compatible checkpoints/configs.
 """
 from __future__ import annotations
 
@@ -38,7 +38,8 @@ import torch
 from torch.utils.data import DataLoader, Subset
 from tqdm.auto import tqdm
 
-from dataset import HealthDataset, collate_fn
+from dataset import HealthDataset
+from eval_data import load_sequence_eval_dataset, sequence_eval_collate_fn
 from models import DeepHealth
 from readouts import build_readout
 from targets import PAD_IDX, CHECKUP_IDX, NO_EVENT_IDX
@@ -1330,11 +1331,14 @@ def main() -> None:
         torch.backends.cudnn.benchmark = True
 
     print("Loading dataset...")
-    dataset = HealthDataset(
+    dataset = load_sequence_eval_dataset(
+        model_target_mode=model_target_mode,
         data_prefix=data_prefix,
         labels_file=labels_file,
         no_event_interval_years=no_event_interval_years,
         include_no_event_in_uts_target=include_no_event,
+        min_history_events=int(cfg.get("all_future_min_history_events", 1)),
+        min_future_events=int(cfg.get("all_future_min_future_events", 1)),
         extra_info_types=parse_int_list(cfg.get("extra_info_types", None)),
     )
     validate_dataset_metadata(dataset, cfg)
@@ -1346,7 +1350,7 @@ def main() -> None:
         subset,
         batch_size=int(cfg_get(args, cfg, "batch_size", 128)),
         shuffle=False,
-        collate_fn=collate_fn,
+        collate_fn=sequence_eval_collate_fn,
         num_workers=int(cfg_get(args, cfg, "num_workers", 4)),
         pin_memory=device.type == "cuda",
         persistent_workers=int(cfg_get(args, cfg, "num_workers", 4)) > 0,