I just started exploring the idea for adding live-stacking support in Ekos. Should this in Ekos or a different process altogether? Should we just use Siril or ASTAP instead instead of wasting resources on this?

Here is a very preliminary Lucid diagram for how I envision this could work. An image of the chart is also attached.

Please feel free to voice your opinion, especially if you're a developer who can contribute to making this happen. What can be done to expedite this? any code that could be re-used from 3rd party (Open source) sources? Any issues with the process as outlined?

I'd say outside of ekos, just to keep things simple, modular, and easy to fix/develop.

On my machine, siril is about 4.5MB and maps 181 libraries. Kstars maps 151 libraries. From the set 251 different libraries these programs use, there are 77 libraries in common, 174 libraries used by one of kstars or siril, and not the other. I haven't tried to profile Siril's livestacking memory use, but in stacking regular images it'll happily use all the memory you can give it. I fear that this would not play well with kstars.

I much prefer the image quality produced by Siril, but AstroLiveStacker has been fun to play with too.
indilib.org/forum/development/5204-als-astro-live-stacker.html

It appears currently we have these open source livestackers:
1. ASTAP
2. Siril
3. AstroLiveStacker

Any good comparison of the three? maybe there are more?

A couple more:
* github.com/art-den/astra_lite
* github.com/rickbassham/livestack

It would be nice if Live Stacking would work on KStars for Windows as well.
I have no opinion on how to implement Live Stacking at this time.

I have been looking for a suitable self-contained live stacking program that is easy to plug in. The criteria that I used are:

1) Stack from a folder. This is useful for cases when you don't have a camera hooked up and want to be as independent as possible from the embedded target (so long as you can export a folder to the host).
2) Decent and easy to use graphics to stretch the stacked image.
3) Speed, responsive.
4) Retain stretching parameters when a new sub comes in. Of course, this only matters with decent graphics.
5) Free.
6) Cross-platform. It's nice if you can run it on host and target. However, it may be better to run it on a host to not affect the real-time target behavior.
7) Scriptable (to customize preprocessing and stretching features).

I scored ASTAP, Siril, DSS Live, AstroToaster, ASIStudio, StellarMate and SharpCap on these criteria but won't present a table here - I would have to double-check for liability reasons.
The hardest criteria turned out to be the combination of (2,4) because I found only one suitable candidate suitable, maybe two (one that I have not tried myself but that I heard was good). Without it, it's worthless regardless of how much effort was put into the rest.

Just some usability criteria from a user's point of view, HTH.