build(train): add python-dotenv for .env credential loading

[tkgstrator]

What

• Add `python-dotenv` to dependencies and call `load_dotenv()` at the start of `main()` in `train_board_ocr`, `train_piece`, and `train_detector`.
• Repository-root `.env` now feeds `HF_TOKEN`, `WANDB_API_KEY`, `CF_ACCESS_CLIENT_ID`, `CF_ACCESS_CLIENT_SECRET` into `os.environ` before HuggingFace or wandb resolve credentials.

Why

• Silences `Warning: You are sending unauthenticated requests to the HF Hub` without having to `export HF_TOKEN=…` in every shell.
• `load_dotenv()` does not override existing `os.environ`, so devcontainer-exported values still win. Missing `.env` is a no-op — nothing to break on machines that already export the tokens.
• `.env` is already in `.gitignore`, so no secret-leak risk.

Version

Bumped to `v0.3.1` (patch: new user-visible capability, no breaking change).

Test plan

• With no `.env` and no exported tokens: HF still warns as before (baseline preserved)
• With `.env` containing `HF_TOKEN=hf_xxx`: warning goes away and no exports required
• With both a `.env` HF_TOKEN and a shell-exported HF_TOKEN: shell value wins (verify via a debug print)

🤖 Generated with Claude Code