refactor: 重构打包模块，新增 BFD/FFD/Greedy 三种 bin-packing 算法，默认 BFD

- 将 pipeline/packing.py 拆分为 packing/ 子包 (base/stream/binpack)
- 新增 BfdPacker(默认)/FfDPacker/GreedyPacker，移除 StreamingPacker
- 超长序列直接截断至 pack_size
- group_size 语义改为"每 N 个 chunk 合并为一块"，默认 1000
- 新增 AutoTokenizer.token_to_id()，修复 ChatML 中 hacky 的 nl_id 获取
- pad_value 默认改为 2(pad_token_id)，position_ids pad=0, loss_mask pad=False
- 新增 position_ids 打包后归零一致性测试
- scripts/cache_h5.py 新增 --pack-algo 参数

pipeline/__init__.py

@@ -28,7 +28,13 @@ Usage::
 from pipeline.pipeline import Pipeline, PipelineConfig, Stage, TransformStage
 from pipeline.tokenize import AutoTokenizer, ChatTemplate, train_bpe_tokenizer
 from pipeline.text import TextNormalizer
-from pipeline.packing import SequencePacker
+from pipeline.packing import (
+    GreedyPacker,
+    FfDPacker,
+    BfdPacker,
+    BasePacker,
+    pack_tensors,
+)
 
 # I/O module
 from pipeline.io import FileScanner, HDF5Handler, export_dataset, cache_jsonl
@@ -70,7 +76,11 @@ __all__ = [
     "train_bpe_tokenizer",
     # Text processing
     "TextNormalizer",
-    "SequencePacker",
+    "GreedyPacker",
+    "FfDPacker",
+    "BfdPacker",
+    "BasePacker",
+    "pack_tensors",
     # I/O
     "FileScanner",
     "HDF5Handler",