Keep in mind that enterprise/datacenter grade SSDs with a build in powerloss protection are recommended for ZFS. Most consumer SSDs can't handle sync writes well, might corrupt both SSDs at the same time so mirroring won't help on an power outage and might not survive that long (years or just months) because with all that write amplification of virtualization, ZFS with coW, mixed blocksizes, nested filesystems, ...
Also make sure you don't get HDDs that use SMR if you want to use them with ZFS (or any other workloads other then doing some basic office stuff) as the write performance will be terrible and the IO delay that high that ZFS might think the disks are dead because they are too slow to answer in time.
And a Ryzen 5 3600 isn't that bad when it comes to power consumption. TDP is more about cooling not power efficiency. Lets for example say (no real numbers) that the i3-10100TE is rated for 35W and the Ryzen 3600 for 65W. That means with disabled boosting the Ryzen should need 2 times the cooling capabilities because it can consume double the power. But when idling both CPU consume really low power (lets say 10W).
Lets say the Ryzen is double as fast as the i3 so it might finish a job in half the time, so for a task the i3 would need 2 minutes the Ryzen could do it in 1 minute. That also means that the i3 will consumer 35W for 2 minutes while the ryzen will use 65W over 1 minute and then only 10W the other minute because it can clock down and send cores to sleep earlier while the i3 is still working at full load. So in average the Ryzen will also only consume 37,5W over those 2 minutes ((65W+10W) / 2 minutes) so its not really more then the i3 but its a way better experience because the job is done in half the time.
So less TDP isn't always better. Its true that lower clocks are more power efficient but it isn't that much more power consumption to run a CPU with moderate clocks (you could also underclock the ryzen to be more power efficient) and you might get a way better user experience with just a bit more power consumption.
And I personally wouldn't buy a ITX/uATX board again for a server. As soon as you start with PCI passthrough you easily run out of PCIe slots. I for example would like to add 7 PCIe cards (2x HBA for TrueNAS, 2x GPU for Win VMs so videoplayback will work without stuttering, 1x Quad Gbit NIC for OPNsense, 1x 10Gbit NIC for fast networking, 1x PCIe to M.2 card because only one of M.2 slots of consumer mainboards is usually connected to the CPU so the other M.2 slots got low performance...some people also need USB cards because emulated USB passthrough isn't that great, sat receiver cards for DVR and so on) in my homeserver but my uATX board only got 3 slots. So if you got the space a ATX board with up to 7 slots is way more flexible.
And if you want to use PCI passthrough you also need to check which mainboard to buy as not all mainboards got useful IOMMU grouping. And ECC support for consumer boards is rare too.
