Somewhat of a necropost (apologies), but as a long-time member of the Discourse community, and a Discourse plugin developer, this is an interesting discussion.
neptronix I understand where you're coming from. The pain of managing a plugin for a changing framework is real. It's more laborious to test your code properly, you lack control and your roadmap is contingent on someone else's roadmap. There is maintenance overhead that is often under-appreciated. If your timeline is years, then this is a real consideration.
That said, after building and maintaining plugins in production for many communties for a number of years now, I think there are ways to significantly ameliorate this pain to the point that I wonder whether it should be the crux of your decision. The ways I'm thinking about include release pipelines, controlled updates, quality code and engaged maintainers amongst others. Also, it really depends on the kinds of customisations you're talking about. If it's an ambitious extension / plugin that goes significantly beyond what's anticipated or explicitly supported by the framework, this question is much more relevant. If you're talking about an anticipated / conservative extension, the importance of this quesiton is reduced.
From the sounds of it Flarum may well check a number of those boxes, particularly for your intended use (i.e relatively standard, non-ambitious extensions). There's a long history around this question in Discourse, however I would point out that it has had stable client and server-side APIs for some time and has recently released a more mature plugin version management system (more mature than the previous version, which you'll see reference to in the discussion): https://meta.discourse.org/t/pinning-plugin-and-theme-versions-for-older-discourse-installs/156971. There is still, of course, stability issues for ambitious plugins with functionality outside of those APIs.
As a community plugin developer I like the sound of where the Flarum team's head is at regarding the place and role of extensions. I, of course, like Discourse and would advocate for it. But that is largely a function of familiarity and vested interests. No one is really "objective" on these kind of questions.
On the Vanilla front, I see there is some chatter about their development going private which would be somewhat concerning if I were thinking of going down that path: https://open.vanillaforums.com/discussion/38318/has-vanilla-development-gone-private.