There are already a bunch of these. gtkcord (disset) comes to mind for an open-source native client: https://github.com/diamondburned/dissentThey seem to be most popular with the open source crowd. I think you could gain traction with the "normal" users by also including many of the gaming features that other native clients lack (game streaming, voice/video chat, soundboards, the lot). Also maybe consider actually supporting all of Discord's chat features, even the ones you don't personally use; things like forums, image boards, announcement channels, and all the other less-obvious Discord features are rarely implemented by third party clients and make for a pretty limited experience if you're in a server that uses them.
I don't think most people on Discord really care about the 2 second throbber on startup or the 500MiB of RAM that Discord wastes when all other applications on the system do the same. Discord isn't the fastest application and it's wasting resources, but it's not too slow. However, the same way Kunlun Tech Co., Ltd. has managed to sell Opera as a "gamer" browser with a non-insignificant market share, I think you could get traction by highlighting the performance aspects such a streamlined client permits.
FWIW Discord's TOS do not allow for third-party clients, and they have banned people using them (though usually only because the clients interacted with the APIs regarding friends, because those can be abused by spambots and scammers).
> So are Nitro features
I'd tread very carefully with supporting Nitro features in a custom client. If their API has endpoints for them I think you may be in the clear, but if there's even a remote chance that you'll be giving free users paid features, Discord will take notice quickly. Part of Discord's income is derived from spamming people with Discord Nitro promotions. If your alternative clients will get big enough without these, they'll probably notice.