I agree with pretty much everything you said. There are a lot of interesting programmatic and design questions that come up. The ONE place I have to nitpick though is that police departments work on shoestring budgets. Remember, if IA is doing its job, they'll have a healthy number of officers who are getting disciplined in one way or another, OR who are being investigated. If the cameras do their job, that number should plummet. Based on case studies in Rialto, complaints against the department dropped by 88% with cameras being worn. Police use of force dropped by 60%. Each of these have direct budgetary impacts on the police force. Considering a single successful lawsuit against the police force for an excessive force complaint can be well over 1m dollars, I'm pretty sure the program will pay for itself relatively quickly. Even without the lawsuits, the lack of having tons of false allegations against the police to pursue frees up a lot of resources. I'd wager good money that the program ultimately will pay for itself.
The more interesting quandry is how to section up the video, provide metadata to mark off what crimes/crime scenes each video snippet is associated with. Do you only retain video from crime scenes that are still within statutes of limitations? What if the officer used force during that encounter, do you purge the video anyway? What if the officer has further issues with excessive force down the road, how do you make sure the video hasn't been deleted? Do you slice and dice the video into chunks for each case? I think there's a lot of interesting questions here. Thus far, departments who have implemented it have chosen to just keep the data indefinitely.