I don't think a Pi5 is in the works.
Given the timelines involved with designing and validating new hardware, especially when it involves chip design, we can be very certain that Pi 5 development is quite far along already. I know a Pi employee (jamesh, on the forums) mentioned already ~1 year before the Pi 4 was released that they had a good understanding of what the Pi 5 would be. I'd guess that at this point we're at most ~1.5 years from a Pi 5 release. I would hope to see it already this coming summer, but that may be a bit too optimistic.
Of course, the Pi 4 has still not peaked, since we're still waiting for Pi OS 64-bit, the full KMS video driver and Vulkan support. Personally, though, I believe the Pi 5 will in large part compatible with all the SW development work that's going on now, i.e. they will most likely keep the GPU very similar (at least from an architectural perspective).
It's pretty hard to speculate about what the Pi 5 will look like, but I mostly agree with @george-spiggott. Regarding using A75 cores, power consumption of the Pi 4 is already pretty much at the limit for what's practical when it comes to passive cooling. It's unlikely that they'd increase it further. This makes it challenging to provide a generational performance leap without going down to more dense manufacturing processes (such as 20nm or 16nm). So, I'd expect a process upgrade for the Pi 5. I also believe that at 16nm (and maybe also 20nm?) they will need to make some provisions to be able to keep the 3.3V I/O on the 40-pin header, since the silicon is not 3.3V tolerant. This will add cost and complexity.
My hope for the next Pi is that they're able to alleviate some bottlenecks in the design. It seems like memory bandwidth is lower than it should be, possibly because of internal bus limitations. My guess is that this is one of the reasons the performance of N64 emulation is still quite bad, despite seemingly not being particularly CPU or GPU limited on the Pi 4.