Expand 3D captcha into three subtypes: 3d_text, 3d_rotate, 3d_slider

Split the single "3d" captcha type into three independent expert models:
- 3d_text: 3D perspective text OCR (renamed from old "3d", CTC-based ThreeDCNN)
- 3d_rotate: rotation angle regression (new RegressionCNN, circular loss)
- 3d_slider: slider offset regression (new RegressionCNN, SmoothL1 loss)

CAPTCHA_TYPES expanded from 3 to 5 classes. Classifier samples updated
to 50000 (10000 per class). New generators, model, dataset, training
utilities, and full pipeline/export/CLI support for all subtypes.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

training/train_3d_text.py

@@ -0,0 +1,40 @@
+"""
+训练 3D 立体文字验证码识别模型 (ThreeDCNN)
+
+用法: python -m training.train_3d_text
+"""
+
+from config import (
+    THREED_CHARS,
+    IMAGE_SIZE,
+    SYNTHETIC_3D_TEXT_DIR,
+    REAL_3D_TEXT_DIR,
+)
+from generators.threed_gen import ThreeDCaptchaGenerator
+from models.threed_cnn import ThreeDCNN
+from training.train_utils import train_ctc_model
+
+
+def main():
+    img_h, img_w = IMAGE_SIZE["3d_text"]
+    model = ThreeDCNN(chars=THREED_CHARS, img_h=img_h, img_w=img_w)
+
+    print("=" * 60)
+    print("训练 3D 立体文字验证码识别模型 (ThreeDCNN)")
+    print(f"  字符集: {THREED_CHARS}  ({len(THREED_CHARS)} 字符)")
+    print(f"  输入尺寸: {img_h}×{img_w}")
+    print("=" * 60)
+
+    train_ctc_model(
+        model_name="threed_text",
+        model=model,
+        chars=THREED_CHARS,
+        synthetic_dir=SYNTHETIC_3D_TEXT_DIR,
+        real_dir=REAL_3D_TEXT_DIR,
+        generator_cls=ThreeDCaptchaGenerator,
+        config_key="3d_text",
+    )
+
+
+if __name__ == "__main__":
+    main()