I'm happily running 4 sticks of Crucial BL16G36C16U4B.M16FE1 with its default 3600 CL16 XMP profile (16-18-18-38) and a 5950X-B2. Memory has been stable on DOCP defaults with BIOS 0801, 0904, 1002, and 1101. I can't remember for sure, but it may have been stable all the way back to BIOS 0211, which was the factory BIOS for my board; I just know it's definitely been happy for 0801 and later on both default JEDEC speed and default DOCP speed. That Crucial part is JEDEC 2666 CL19/20 Micron RAM. Is your RAM listed as "1,2,4" on the socket support column of the QVL? 4 DIMM configs always push the limits harder than 2, and only the entries listed as 1,2,4 are validated for 4 DIMMs. That's not to say that only the validated 4 DIMM configs will work, just they will improve your chances of success.
You can use M.2_2 and M.2_3, just have to run the GPU slot as x8. It gives you the ability to run a blazingly fast RAID 0 with around 21Gbytes/s bandwidth when combined with M.2_1. The DIMM.2 slots are via the chipset, so limited by the X570 platform's x4 chipset link to roughly 7Gbytes/s shared between them and all other chipset devices. DIMM.2 is useful for capacity while keeping the GPU running at x16. It gives you choices about where you want to place the performance, and x8 isn't necessarily a huge performance hit for RTX 3000 and RX 6000 GPUs (which were the latest for the tail end of the X570's reign). Some games/applications do need the x16 bandwidth, but many are really just fine on x8. If you're someone who needs every last bit of framerate, x8 may steal a little. Ryzen 5000 only has a total of 24 PCIe lanes to play with, 4 of those always have to go to the chipset, leaving only 20 lanes direct to the CPU; all of the Crosshair boards do something creative with those 20 lanes to give you choices beyond just a single x16 for GPU and a single x4 for NVMe (AMD's platform design for AM4 / X570 is essentially just 16+4+chipset).
