Fix ffmpeg install on macos

[ejm714]

• install ffmpeg with brew on mac
• set macos version for proper densepose installation
• fix failing test due to arbitrary image -> video fps
• unpin botocore due to uv fix Avoid preferring constrained over unconstrained packages astral-sh/uv#3148

Closes #315
Closes #326
Closes #327
Closes #328

[netlify]

✅ Deploy Preview for silly-keller-664934 ready!

Name	Link
🔨 Latest commit	9cee50a
🔍 Latest deploy log	https://app.netlify.com/sites/silly-keller-664934/deploys/6646be1d6337f20008aa5130
😎 Deploy Preview	https://deploy-preview-329--silly-keller-664934.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[github-actions]

🚀 Deployed on https://deploy-preview-329--silly-keller-664934.netlify.app

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 87.4%. Comparing base (70e79f2) to head (9cee50a).

Additional details and impacted files

@@          Coverage Diff           @@
##           master    #329   +/-   ##
======================================
  Coverage    87.4%   87.4%           
======================================
  Files          26      26           
  Lines        2191    2191           
======================================
  Hits         1915    1915           
  Misses        276     276           

[ejm714]

It seems like something funky was happening in the creation of the video from individual images (which is not our standard workflow, the goal was just to create a video with known frames with objects). We separately test MegadetectorLite just on the stack of 10 frames and get the right detections out.

zamba/tests/test_megadetector_lite_yolox.py

Lines 70 to 90 in 70e79f2

	def test_detect_video(mdlite, dog):
	total_frames = 10
	object_frame_indices = [0, 3, 4, 6]
	video = np.zeros([total_frames] + list(dog.size[::-1]) + [3], dtype=np.uint8)
	video[object_frame_indices] = np.array(dog)
	mdlite = MegadetectorLiteYoloX()
	detections = mdlite.detect_video(video)
	# Check that we detected the correct number of frames
	assert sum(len(score) > 0 for _, score in detections) == len(object_frame_indices)
	for frame_index, (frame, (_, score)) in enumerate(zip(video, detections)):
	if len(score) > 0:
	# Frame index is in intended frame indices
	assert frame_index in object_frame_indices
	# No blank frames were selected
	assert not (frame == np.zeros(frame.shape, dtype=np.uint8)).all()

I can't explain why but any frame rate other than 60 makes this test pass 🤷‍♀️ I use 30 here since that's typically what we see in videos coming in

[ejm714]

the default is a confidence of 0.25 so this isn't changing the underlying behavior