My opinion on K vs. KF:
KF originated so Intel could sell more units during production not meeting demand. They took CPU’s that would normally fail Q/C due to a bad GPU and then disabled it and sell it as a KF. It’s not an evil thing to do - most six core chips for example were octa core chips with one or two bad cores disabled. Same idea, applied differently.
With the onboard GPU disabled you may have a marginal increase in available thermal headroom for overclocking but I doubt it’s enough to really care about.
The fact about having onboard as a backup during GPU failure? I’d keep it just for that! If you have a second PC for general purpose use, then it doesn’t matter. If you are an one machine kinda guy, have that onboard there, especially if the budget for a replacement GPU won’t be easy to come by for a while (don’t bet on warranty).
A K can also be used without a GPU intentionally, if you hand down old hardware to spouse/kids. If they don’t need a dedicated GPU, removing it saves on power usage. I built my dad’s new tower with a 9th gen I5 and no GPU - his office is no longer a boiler room! (Came from an i7-920 gen1 / my old 560TI).
Yup, this is the single best bang for buck on new boards.
AMD has boards with PCIe 4.0 - does anyone know how close we get to maxing out PCIe3 x16 with current (30xx) GPUs? Last I heard was around the 10xx cards and they weren’t close, I recall debates if x8 mode was really a penalty or not (for cases where you have other PCIe cards attached).
For the Intel 10th gen everything got hyperthreading and therefore jumped up the scale a bit. 9900k is 8C/16T and the 9700K is 8C/8T. Go to tenth gen and the 10900 is 10C/20T while the 10700 is 8C/16T. So the new I7 is the old I9 essentially. The I5 doesn’t change up to the I7 but it is now 6C/12T instead of 6/6 so that’s still a big jump.
Schurem is right, there is a too fast. We’ll still benefit from quantity before quality so start at 32GB. Most “on the box” speeds are actually XMP profiles. The CPU and Motherboard dictate base speed. 9th gen was 2133 to 2666. 10th gen Intel is around 2933. Someone else can quote AMD haha! 
You also can lose with XMP - you may get faster clocks but the latencies also go up which may be worse. That’s an application-specific research project. Last I read, added latency was bad for DCS but these newer 3000mhz plus profiles may get different results.
If you don’t overclock, you can still buy OC capable RAM and just let it run at the base speed, just buy whatever is more economical.