l2arc is just a read cache. by definition if it fails data is not lost.
in contrast i’d discourage from making a metadata device unless you’re mirroring it - one slice of failure, metadata is not a nice-to-have. Seemingly it can be worth it, but unless there’s a serious workload, that may still not be enough reason.
One thing to consider, is how large do you consider the zpool to ever grow to. if that’s 8 bays, you might not want a pool so wide as to be only a z1. If so you might want to make it a z2 now (meaning 4th disk). That won’t be the most efficient or performant configuration now, but zfs expansion can grow a raidz pool width but not change the parity.
if you’re DIY’ing your server rather than through something like TrueNAS, careful to initialise the pool at 4k blocks (ashift=12) (also often advised is to apply atime=off, that’s at least configurable later, like sync, but ashift is not).
The nice thing is you can make your own choices for the characteristics you want. best of luck