From those choices I'd say focus on features and keep AGS3 bug fixing legacy support for a few years.
I'd say don't be afraid to demolish as much as you have from the old code (or even start from scratch taking advantage of newer technologies) and focus on easy of portability to newer platforms.
It's been a while since I used AGS, but I noticed that most of my user base plays my games on mobile.
And every now and then I talk to fellow developers talking about picking engines and a lot of the time they skip over AGS because of this same reason.
Edit:
As for the legacy AGS engine, I'd say to wrap it around some ScummVM-like UI to help people that use it to run older games to have a tool just for that.
I'd say don't be afraid to demolish as much as you have from the old code (or even start from scratch taking advantage of newer technologies) and focus on easy of portability to newer platforms.
It's been a while since I used AGS, but I noticed that most of my user base plays my games on mobile.
And every now and then I talk to fellow developers talking about picking engines and a lot of the time they skip over AGS because of this same reason.
Edit:
As for the legacy AGS engine, I'd say to wrap it around some ScummVM-like UI to help people that use it to run older games to have a tool just for that.