They're fabbing with TSMC, which has a really established history in high yields. Just think about the A13/A14 CPUs, which are probably the most voluminous single CPU models every fabricated by a company, yet have never been subject to binning to other lower or higher end models. As the chips get more complicated naturally there will be more fabrication issues, so I expect we'll see more binning, but it doesn't look like there's much cause for concern.
[1]: https://www.anandtech.com/show/16028/better-yield-on-5nm-tha...
A quick look in ifixit's HomePod teardown (2018) reveals an Apple A8 processor, last seen in the iPhone 6 (released 2014): https://www.ifixit.com/Teardown/HomePod+Teardown/103133
I wonder how long they manufacture their processor lines for.
Yes, you bin the whole SoC; but all the performance differences are expected to be concentrated in the CPU and GPU. Expect a high bin where all cores and all GPU subprocessors are working, a medium bin where 1 or 2 CPU cores and 10% of the GPU is either missing or running at a lower frequencey, and a value bin where perhaps half the CPU cores and up to half the GPU are downgraded.
My guess is that any multivariate binning will be lowest-common denominator, e.g. an 8-performance core / 12-GPU core part with a single, second-tier bin of 6 performance cores and 10 GPU cores.