I simply counted the source lines in s6[10k], skalibs[4k] and execline[16k], all of which were needed to compile the s6+skalibs+execline suite [I just did this]. If only 1k lines are really needed for a supervisor, then there must have been a way to just limit the attack surface to just this subset.
In his writings on init systems, Steve Litt covered all these supervisors, so maybe I will just ask for some advice on these matters [one day].
There is a school of thought that fail-safe supervisors for daemons are not just unnecessary, but positively dangerous. I have no experience managing major server systems, so just don't know. In the year or so that I have been using sinit, I have not seen any of my daemons die [but I only have udevd, ntpd, klogd, syslogd, cupsd].
It was long ago clear that the decisions to adopt systemd were based on political and commercial rather that technical reasons. An in-depth study of systemd may one day show it to be just great, but I doubt it. In the meantime, it is best to search for viable alternatives [of which there are many].
My main concern these days is at the eye-watering scale of software "bloat". I estimate that of the 15 million sloc in the Linux kernel only [only!] some 500 thousand are used in a typical deployment [compare 260 mb of kernel modules to 3-6 mb kernel binary]. Not to mention the billion sloc in the Linux userland. It is not just the scale of all this, but the dependency nightmare: there are many hundreds of libraries in a minimal Devuan install.