feat(engine): 训练评估循环、CenterNet 损失与指标

- trainer: AMP、梯度裁剪、cosine/step 调度、last/best 检查点，按 monitor 指标挑选 best
- losses: penalty-reduced focal loss（heatmap）+ 掩码 L1（wh/offset）+ 分类损失
  （多类 BCE / 单类 CE，随 label_mode 切换）
- metrics: 定位 IoU 匹配下的 P/R/F1 与 mean IoU；分类多标签 micro-F1 与单标签 accuracy，
  两种模式都输出 cls_f1，故 train.monitor 无需跟着改
- optim: BatchNorm 与 bias 不做 weight decay
- visualizer: SwanLab 记录损失/学习率/验证指标/显存峰值，未安装或离线时降级为警告

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

engine/__init__.py

@@ -0,0 +1,17 @@
+"""训练 / 评估流程。"""
+
+from engine.checkpoint import load_checkpoint, save_checkpoint
+from engine.evaluator import Evaluator
+from engine.losses import DetectionLoss
+from engine.optim import build_optimizer, build_scheduler
+from engine.trainer import Trainer
+
+__all__ = [
+    "Trainer",
+    "Evaluator",
+    "DetectionLoss",
+    "build_optimizer",
+    "build_scheduler",
+    "save_checkpoint",
+    "load_checkpoint",
+]