A few notes about analytics and why they prefer MJPEG. While H.264/265 is awesome with compression for storage and bandwidth savings its a huge resource hog on the VMS to decode and analyze video. Most VMS platforms and Analytics Serves prefer to preform analytics analysis using MJPEG at lower frame rates to conserve CPU resources, it takes les CPU to decode the MJPEG Stream and then throw it away while writing the H.264 to disk for playback.
While some do it better than others we see the trend of offloading resources to the edge to free up more VMS/Analytics CPU processing power, things like motion on the edge, analytics embedded into camera/edge devices etc.
With regards to your other issues of the camera settings changing see if the knock off camera (sorry it sounds like that what it is you selected) has the ability to capture/create log files so you can try and get help reviewing them. Wait it does not sound like you will get anywhere w/log review since the camera does not have the advertised features as stated so I suspect the support will be the same. Do the same for Milestone, have the logs reviewed for the server/camera that is in question, the logs will tell you the stream profile and when it was changed. The VMS sends the profile and the cameras then send those parameters back. That is assuming you are using a cameras that has a driver in Milestone, they don't do well with generic RTSP streaming. If you are using a generic RTSP stream that is likely the problem but log will tell for sure. Even ONVIF has its challenges with IP cameras that claim to be compliant and really are not even close.
One final note, I spent almost the last ten years doing analytics for a manufacturer, you need to make sure you don't cut corners on the video source that will feed the analytic engine because the analysis is only as good as the source its analyzing . That being said LPR is even more demanding of a purpose built device, don't try and take the "best in class" (aka cheaper) device to capture plates and then complain when it does not work right. The money you saved in the long run you will eat trying to make it right for the customer. Great example is end user complains they can only see the plate and not the vehicle or the entire scene of the area of interest, well this is an LPR application right? Goal is to capture plates so the LPR analytic can index, search and alert. If you need the scene overview then install a camera that is purpose built for those conditions along side the LPR camera.
If LPR or analytics are not your forte` then punt when asked or at least make sure you consult with the analytic manufacturer and LISTEN to what the share with you, their knowledge is specific to their product and how to make it work. I would be that if you likely told them you were using a non-purpose built camera they probably told you something like "don't call me when it does not work".