Johnnie Hancock wrote a great article about this topic:
https://www.electronicdesign.com/home/article/21200854/overcoming-overdrive-recovery
Yes, Johnnie Hancock just states the obvious: do not overdrive your oscilloscope frontend. Just don't do it, no matter what use case you come up with. Just bcause a use case exists, that's no excuse to abuse a (maybe inadequate) tool.
Especially with modern high resolution DSOs, we do have alternatives. Even when we really need that much of a dynamic range, because the desire to analyze a small detail in a high-amplitude waveform, there would be alternative approaches with modern high resolution DSOs, so that we don't need to overdrive the oscilloscope frontend and risk overload (recovery) distortions.
In my thread I've demonstrated multiple times how vertical zoom in combination with certain math functions (I've preferred to use Average on periodic signals, but ERES could be used as well) can increase the resolution up to 16 bits and reduce the noise at the same time, so that high zoom factors become perfectly usable - that's pretty much what Johnnie Hancock recommends in his article.
Johnnie Hancock only talked about the overload recovery times of the semiconductor amplifiers, where semiconductor components require a certain time to recover from saturation. Of course we have to deal with that in Siglent oscilloscopes as well, namely in the integrated PGAs from a reputable major semiconductor manufacturer, but there recovery times are more in the below-one-nanosecond region and certinly not much of a concern for a 200 MHz DSO.
What Johnnie Hancock forgot to mention, is the much more important problematic overload recovery behaviour of the split path input buffers in DSO frontends, as they are universally used for general purpose oscilloscopes. These offer reasonably stable DC offset even with high bandwidth frontends. While the LF-path will clip cleanly and its recovery time doesn't matter because the signals here won't exceed one kilohertz or so anyway, the HF-path differentiates the signal instead of clean clipping, hence throwing it out of sync with the LF path. Here we are not talking about nano- but milliseconds!
Btw: the higher the offset compensation capability of a scope frontend with split path input buffer, the more potential problems might occur.