Revisit or reconsider centroid change monitoring

[klauer]

• Goal: Single frame drops should not affect overall monitoring results (assuming a valid frame comes in under the maximum time threshold)
• We made some changes at the last minute while deploying, tweaking the algorithm to not check thresholds on a per-frame basis

Possibilities I can think of at the moment:

• The double-threshold value can probably be collapsed back into the original single-threshold check
• Alternatively, we can restructure the algorithm to somehow look at the bigger picture (multiple frames, etc.?)

[klauer]

Reconsider centroid change monitoring altogether -> the image sum is preferable for verification overall. Closing this and opening a new issue for replacing the mechanism.