Another reason is that ME (and PSP) are used to assert "security" of "Protected Media Path", which is part of streaming services DRM.
It's not the only way to do it, sure, but there's a reason why IBM POWER designs have approximately 2.1x amount of cores that is stated in the spec - pretty much every core has a smaller, simpler one dedicated to keeping the big one running without melting and helping transition to low-power states and back, and there are at least two more cores whose only job is handling some of the early loading of code from flash. Part of why they have those cores is that since POWER8 there was a shift towards more standalone operation without external controller chip (and POWER9 even describes such boot process in manual). For comparison, traditionally the FSP (aka BMC) on IBM POWER systems initialized all of hardware before the CPU would execute first instruction, puppeting the CPU buses through debug interface.
There was a time when HP sold servers that could be up to say, 8 cores but only two were on by default and you cloud license the rest. It was cheaper to shop the hardware and software gate it rather than limit it and have a process in the middle.
Also, customer CPUs are often (due to binning) rebranded Enterprise CPUs that were rejected (or just not purchased) for whatever reason, or vice versa. Easier to build ME on them all and configure it later.
Your ECC memory? It's something Intel just flips on or off depending on how they want to sell the chip - many Core i7s and i9s have it on-die but you can't use it. That's because it might be useful on a Xeon Platinum equivalent.