You can see there that little time was spent reading data, so that explains why repeating the query didn't make it much faster due to caching. The time isn't spent reading data, but doing CPU work on data already in memory.
Agreed. The "perf top" output produced by Kenny definitely indicates that. That's precisely why I concentrated on JIT. Anything else would give us a slim chance of increasing the query speed.
I don't think that that follows at all. Proper index usage can vastly improve CPU-bound performance, not just Disk-bound performance. And that it used to be faster in the recent past suggests that it is certainly possible for it to be faster.
I assume he is on a system/installation without a JIT provider, so even if jit is enabled in the postgresql.conf, it won't be used.