Add support for EIM files with shared memory, add support for setting thresholds

[janjongboom]

Companion to https://github.com/edgeimpulse/edgeimpulse/pull/13597

• Add support for shared memory to exchange features
• Add support for set_threshold (plus example)
• Also clean up / widen the requirements list; so less conflicts when importing this package

[janjongboom]

This should be removed

[janjongboom]

Need to do this because we don't own the shared memory - otherwise will get a warning from the resource tracker that we leave shared memory behind.

[jimbruges]

I have found the for loops at the end of get_features_from_image_with_studio_mode() add quite a lot of latency to inference- switching out for numpy flatten instead speeds things up significantly on slower devices

if is_grayscale:
        resized_img = cv2.cvtColor(resized_img, cv2.COLOR_BGR2GRAY)
        # Use numpy's vectorized operations for feature encoding
        features = (resized_img.astype(np.uint32) * 0x010101).flatten().tolist()
    else:
        # Use numpy's vectorized operations for RGB feature encoding
        pixels = resized_img.astype(np.uint32)
        features = ((pixels[..., 0] << 16) | (pixels[..., 1] << 8) | pixels[..., 2]).flatten().tolist()

    return features, resized_img

[mmajchrzycki]

Indeed, the new implementation is ~10 times faster.

[mmajchrzycki]

I tested it and it looks good. Added some small improvements.

[mmajchrzycki]

Indeed, the new implementation is ~10 times faster.