feat: Add constitutional transforms based on Anthropic Constitutional Classifiers++ paper (#300)

* feat: Add constitutional transforms for AI red teaming

Add constitutional classifiers probing transforms based on Cunningham et al. 2025 paper:
- Reconstruction attacks: code_fragmentation, document_fragmentation, multi_turn_fragmentation
- Obfuscation attacks: metaphor_encoding, riddle_encoding, contextual_substitution, character_separation
- Supports static, LLM-powered, and hybrid transformation modes
- Add comprehensive example notebook demonstrating all transforms with TAP integration
- Strip notebook outputs for clean commit

* fix: Change noqa to nosec for bandit compatibility

Replace # noqa: S311 with # nosec B311 for bandit security scanner compatibility

* fix: Add noqa comments for both ruff and bandit

Add both # noqa: S311 (ruff) and # nosec B311 (bandit) to suppress security warnings for non-cryptographic random usage

Author: rdheekonda

dreadnode/transforms/__init__.py

@@ -12,6 +12,7 @@
 if t.TYPE_CHECKING:
     from dreadnode.transforms import (
         cipher,
+        constitutional,
         encoding,
         image,
         perturbation,
@@ -29,6 +30,7 @@
     "TransformWarning",
     "TransformsLike",
     "cipher",
+    "constitutional",
     "encoding",
     "image",
     "perturbation",
@@ -41,6 +43,7 @@
 
 __lazy_submodules__: list[str] = [
     "cipher",
+    "constitutional",
     "encoding",
     "image",
     "perturbation",