Get something running a haswell cpu or newer.
Given you don't know what the op is planning on running in any great detail, that could be amazing or awful advice. I run an fully loaded R210-II with a Sandybridge 1270v1, 32GB ECC, iDRAC etc, it idles sub 50w in Ubuntu, that's slightly lower than what my Ryzen builds pull idle (other than the A300), and a decent amount lower than my E5 V3/4's, it's also within a few almost watts of the 8th gen i3's/i5's, so efficiency and generations don't always follow. If they want to do hardware transcoding (Plex is mentioned in a subsequent post), then Haswell is still too early, you ideally need Kaby Lake for the UHD630 for HEVC and tone mapping to function properly *if* they have/intend to get PlexPass, which is essentially 8th gen upwards depending on the chip (some of the 7th gen mobiles got Iris Pro 650 and the Skylake Xeon's had P630's which iirc carried the same capabilities). If they want to do realtime gigabit hardware encryption via AES-NI or push 10Gbe then clock speed/core count is still an issue, but randomly throwing out a CPU generation with no context to base it on is just poor.
Op - you're walking a very well worn path, the first thing to do is make sure you're heading in the right direction. Do you understand the cost in terms of power, time and hardware at this point vs where you're heading? For example I am sat next to a silent half rack of servers and drive shelves, why are they silent? Because it's cheaper to run off-site box in a DC with symmetrical gigabit and mount cloud storage via rclone than it is to power an equivalent local server with way less local storage. That also ignores the up-front costs to buy, maintain and manage local storage and hardware along with the heat/noise and deal with backups. Ironically for a lot of what I do, it can actually be quicker (you loose in absolutely latency terms, but gain in terms of spinning up a local platter). OK if you're stuck on ADSL or other slow broadband, then this path isn't for you, but on an OK FTTC line or better, it works very well.
If you decide you prefer local, power needs to be considered, too often people buy what is essentially e-waste because they think they need a 'server'. I wouldn't personally touch anything lower than a 7th gen Xeon/mobile CPU with an appropriate iGPU (P630/IP 650), but generally you'll get a lot better value and more capable platform going with an 8th gen i3 at this point, they're efficient, had 4c and were largely comparable to the previous gen i5's and relatively modern HD630 iGPU's so you have tone mapping and full HEVC support, they're also available for next to nothing now. A simple ATX/MATX board and HBA (if required) along with the appropriate SATA break out cable will give you sufficient options to add local drives/NVMe for cache. Case wise whatever has enough room for what you need. Avoid high speed (10-15K SAS) and/or low density drives, avoid shingled drives, also consider that drive density may mean the cost per TB is higher, but that now needs to be balanced with w/TB and higher density - while perhaps not making absolute up-front financial sense - can make much more longer term sense.
In terms of HA, the Pi3 and 4 are decent hosts, much like virtualising a router, ask yourself what the impact of you taking down the host will be for reboots and upgrades or replacing a drive and if you (and potentially other users) are going to be OK with that or you trouble shooting an issue, the Pi4 isn't that power efficient by SBC or even Pi standards, but unless you are Ok with down time, it may be better to keep this bare metal depending on what exactly you are running/doing.
Plex wise, if you do go with local storage, with proper curation and encoding, you shouldn't need to transcode video for local clients, ever. So PlexPass has limited value, if for some reason you are still transcoding (phones/tablets, remote clients), that's got power/heat/noise implications. It'll either be a media issue (make better encoding/bit-rate choices), a connectivity issue (hard wire/make better bit-rate choices) or a client issue (modern clients shouldn't have any issues... Prime Day approaches, that first gen FTV needs to be taken out and shot) or you have no control over the remote client/connectivity (choose better friends/family, or get them better clients).