chore(train): scope wandb project by image size and package version

docs/backbones.md

@@ -364,5 +364,5 @@ BoardOCR では未実装だが、モバイル配信で精度を詰めたい場
 
 各 backbone の val cell_acc / sfen_full_acc / wall-clock は sweep 完了後に W&B で並び、ここに引用予定。
 
-- W&B project: `mito-train-board-ocr`
+- W&B project: `mito-train-board-ocr-w{image_size}-v{package_version}` (例: `mito-train-board-ocr-w224-v0.2.0`)
 - run 名は各 backbone 名そのまま（例: `mobilenet_v3_small`, `convnext_tiny`）

docs/ocr-scaling-outlook.md

@@ -257,7 +257,7 @@ EPOCHS=50 IMAGE_SIZE=288 BACKBONES="mobilenet_v3_small convnext_atto convnext_ti
 RESUME_INCOMPLETE=1 EPOCHS=100 IMAGE_SIZE=288 ./scripts/train_backbones.sh
 ```
 
-W&B project: `mito-train-board-ocr`。今回の sweep run 一覧：
+W&B project: `mito-train-board-ocr`（当時。以降は `mito-train-board-ocr-w{image_size}-v{package_version}` に分割）。今回の sweep run 一覧：
 
 | Backbone | Run ID |
 |---|---|

mito_train/training/train_board_ocr.py

@@ -17,6 +17,7 @@
 from __future__ import annotations
 
 import argparse
+from importlib.metadata import version as _pkg_version
 from pathlib import Path
 
 import torch
@@ -301,8 +302,9 @@ def log_main(msg: str) -> None:
 
     # Only rank 0 talks to W&B; other ranks keep wandb_run=None and log nothing.
     run_name = args.backbone
+    wandb_project = f"mito-train-board-ocr-w{args.image_size}-v{_pkg_version('mito-train')}"
     wandb_run = _init_wandb(
-        project="mito-train-board-ocr",
+        project=wandb_project,
         run_name=run_name,
         run_id=resumed_wandb_id,
         resume="allow" if resumed_wandb_id else None,