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>

generators/threed_gen.py

@@ -45,7 +45,7 @@ class ThreeDCaptchaGenerator(BaseCaptchaGenerator):
         from config import RANDOM_SEED
         super().__init__(seed=seed if seed is not None else RANDOM_SEED)
 
-        self.cfg = GENERATE_CONFIG["3d"]
+        self.cfg = GENERATE_CONFIG["3d_text"]
         self.chars = THREED_CHARS
         self.width, self.height = self.cfg["image_size"]
 
@@ -154,7 +154,7 @@ class ThreeDCaptchaGenerator(BaseCaptchaGenerator):
             char_img = self._perspective_transform(char_img, rng)
 
             # 随机旋转
-            angle = rng.randint(-20, 20)
+            angle = rng.randint(*self.cfg["rotation_range"])
             char_img = char_img.rotate(angle, resample=Image.BICUBIC, expand=True)
 
             # 粘贴到画布