I think it could be both fun and lead to interesting sounds to be able to select a portion of the timeline (and multiple tracks) and convert it into discreet presets that get added back into the timeline. Some quantization options to choose how often to "sample" would determine how accurate it would be. I'm a big fan of letting computers do grunt work! :)
In some ways it's almost the opposite idea as the of the Metasurface by breaking up linear control into points, but I think they might work well together.
On the other side, it'd be interesting to select presets in the timeline and convert them to multiple tracks of automation. The interpolation in this case would be a time saver in terms of having to draw in curves yourself to smooth between points. Linear, exponential, maybe some more exotic options? I feel like this would be a neat alternative to LFOs and a nice way to test different ideas. I'm a big fan of letting computers do grunt work! :)
Interesting, I've been thinking of something similar to your first suggestion, like timeline automation "scenes" or sections.