Not sure how much information they can/will share so I'll provide some reasonable speculation in the meantime. This is all public information and I'm not an insider at nanoleaf or anything, I just also want to see the issue fixed and spent too much time looking into the issue
I think the issue is really on the chip itself. Nanoleaf is using the MG24 chipset from SiLabs, which has their own matter stack that Nanoleaf does not develop, but uses to control the bulb. There's been anecdotal evidence from those with the devboard that this chipset, despite being a flagship from SiLabs, is slow and is having issues. The problems we are seeing (bulbs randomly turning on, thread network stability, etc) are a result of the MG24 chipset crashing. This likely happens more often when the bulbs have increased traffic load. The bulbs are being used as thread routers ("mesh extender" in the nanoleaf app) to pass thread traffic to downstream devices, and when too much traffic comes through, they crash. That's would also explain why there's variation in behavior that people are seeing, if most of your bulbs are close to a border router, then the instability and lights turning on are less prevalent. (This aligns with my own observations that bulbs acting as routers tend to be less stable than those that are not).
Another confounding factor is nanoleaf is also running their proprietary stack on the chipset alongside matter. How much this impacts the issue, I'm not sure.
This puts nanoleaf in an awkward position where they purchased a flagship chipset from a vendor, explicitly designed for IoT thread/matter devices, and it's not working well. It would be financially difficult to issue a hardware revision so their only real choice is to make the MG24 chipset work, hence we have these small optimizations like the .94 update to shift around RAM allocations. I'm guessing that the delay in new updates is nanoleaf being stuck waiting on feedback/support from SiLabs