Add test for AutoModel vs OneVisionEncoderModel output consistency

- Created tests/test_automodel_consistency.py with tests for:
  - Output consistency between AutoModel and OneVisionEncoderModel
  - Flash attention 2 and eager attention implementations
  - Batch input handling
  - bfloat16 dtype consistency
  - Parametrized tests for transformers versions 5.0.0rc1 and 4.53.1

- Fixed duplicate section in pyproject.toml that was preventing pytest from running

Co-authored-by: anxiangsir <31175974+anxiangsir@users.noreply.github.com>

Author: Copilot

pyproject.toml

@@ -47,56 +47,6 @@ exclude = [
     "pages*",
 ]
 
-[build-system]
-requires = ["setuptools>=61.0"]
-build-backend = "setuptools.build_meta"
-
-[project]
-name = "onevision-encoder"
-version = "0.1.0"
-description = "HEVC-Style Vision Transformer for efficient video understanding"
-readme = "README.md"
-license = {text = "Apache-2.0"}
-requires-python = ">=3.8"
-dependencies = [
-    "torch>=2.0.0",
-    "transformers>=4.40.0",
-    "timm>=0.9.0",
-    "numpy",
-    "Pillow",
-]
-
-[project.optional-dependencies]
-train = [
-    "tensorboard",
-    "nvidia-dali-cuda110",
-    "decord",
-]
-dev = [
-    "pytest",
-    "black",
-    "isort",
-]
-
-[tool.setuptools.packages.find]
-where = ["."]
-include = [
-    "onevision_encoder*",
-    "dataloader*",
-    "model_factory*",
-    "training*",
-    "dataset*",
-    "eval_encoder*",
-]
-exclude = [
-    "shells*",
-    "tools*",
-    "llava_next*",
-    "asset*",
-    "pages*",
-]
-
-
 [tool.ruff]
 target-version = "py310"
 line-length = 119