A4-6210 with build in GPU has TDP of 15W. There is no point to optimize anything it is seeping power already. may be try to use tlp to limit max charge level of the battery ( i'm not sure is you laptop is supported). You can play with governors too, but I personaly will not bother. You obviously need multiuser.target but not GUI.
With raid 10 - i would not risk it . With RAID6 ( obviously not on BTRFS) it is fair game if you have solid return policy for drives which are DOA. Go for SAS drives, they are cheaper (but generally hotter and nosier). And look for old-new-stock on specialized sites, no one in enterprise needs say 8tb drives, so they selling them cheap at times.
Get drive, connect , run long smart self-test ( for 18tb it is probably take a day). If it passed you are reasonably sure that it will not die soon. And keep running these test regularly, as soon as they start failing, replace.
Biggest problem will be BW and latency to your lab from the Internet. I would use dedicated hardware and subnet for it. Security wise, if you can make your site 100% static it will help a lot with security. I'm personally set on AWS S3 + CloudFlare combo with static site generator running in my lab. Yes it is not really "self hosted" but worries free solution for me.
both works. Just do not forgot to assign fake serial numbers if you are passing disks. IMHO passing disk will be more performant, or may be just pass HBA controller if other disks are on different controller.
ZFS or BTRF mirror will know which side is at fault due to checksums. I'm more concern about simultaneous falures of two disks. Rebuilding of a RAID puts lots of pressure on remaining disks, so probability that remaining one dies too is much higher. with RAID6 3 disks need to die to lost date, which is less likely but not impossible.
what is wrong with old good LSA 2008 based HBAs with SAS-to-Oculink cable? and for NVMe part, adapter is just port format converter, pick any.