feat: add Qwen3-VL support with image and video token calculation

[thisisiron]

Summary

• Implement Qwen3VLAnalyst with proper video token calculation logic
• Add smart_resize_video() function for Qwen3-VL's video preprocessing
• Add qwen3-vl tests for both image and video token counting

Changes

src/vt_calculator/analysts/analyst.py

• Added Qwen3VLAnalyst class that overrides calculate_video() method
• Qwen3-VL uses different video processing parameters:
  • min_frames=4, max_frames=768
  • Uses video_processor.size for pixel limits instead of image_processor
  • ASSUMED_SOURCE_FPS=24.0 for fallback when video fps is unknown

src/vt_calculator/analysts/tools.py

• Added smart_resize_video() function that handles temporal dimension in video resizing

tests/test_analysts.py

• Added Qwen3VLAnalyst import
• Added qwen3-vl test case for image token counting
• Added qwen3-vl-video test case for video token counting

Test Results

All 8 tests pass:

tests/test_analysts.py::test_analyst_token_count_matches_transformers[qwen2.5-vl] PASSED
tests/test_analysts.py::test_analyst_token_count_matches_transformers[qwen3-vl] PASSED
tests/test_analysts.py::test_analyst_token_count_matches_transformers[internvl3] PASSED
tests/test_analysts.py::test_analyst_token_count_matches_transformers[llava] PASSED
tests/test_analysts.py::test_analyst_token_count_matches_transformers[llava-next] PASSED
tests/test_analysts.py::test_analyst_token_count_matches_transformers[llava-onevision] PASSED
tests/test_analysts.py::test_analyst_video_token_count_matches_transformers[qwen2.5-vl-video] PASSED
tests/test_analysts.py::test_analyst_video_token_count_matches_transformers[qwen3-vl-video] PASSED

[thisisiron]

/style

[github-actions]

Style fixes have been applied and pushed!