Wish you luck. Hope one day instead of Sage that is a bunch of Python glue for completely different and disjointed systems like Maxima, SymPy, Octave, GAP, PARI/GP, etc we will get a one well-integrated (and blazingly fast because written in Rust). I am a big supporter of open-source but after struggling with Sage and Co, decided that buying Wolfram Mathematica license is well-worth it (never regretted since). It would be great if we don't need to do that anymore.
As for the "most wanted" features, I would vote for PDEs, since most other systems (apart from Mathematica itself) do not support them well.
I love the idea of Sage, and think an open-source state of the art CAS is essential in this day and age, but in trying to master it I got sick of Python.
I truly, honestly loathe the way Python handles for symbolic computation, and digging down I traced my disdain all the way to the fundamental object model of Python; so there is no way some surface-level modification will work for me.
On the one hand, it allows the sloppy "integration" of various systems (like you mention). On the other hand, it is not a real integration, just a patchwork of the worst kind. You never know what kind of interface you will face next, there is no coherence among tools, everything is its own world and you need to translate manually between them. Assuming you know the interfaces beforehand. If not... good luck.