sysdata: streamline metric collection

[lasers]

Merge sysdata: drop legacy first #2348.

---

(qol cleanup) OK. I wanted to do this for some time... to move them into functions and keep updating with returned dicts in the loop, but codex says it's better to mutate everything instead... that way we don't have any really short-lived dict initialization and stuff like that. OK, we'll mutate.

I made minor change to arc_get_size... assuming that zfs, loaded on startup in /proc or not, will either stay there indefinitely or not. Users without zfs gets 0... and no zfs regex is needed because the line looks very easy to parse.

Moved lot of things to post_config_hook too so we can stick with minimal functions including optional last one max_used_percent.

In short, sysdata can be 1-20% faster than before compared to master. It is a wash with all placeholders. The expensive one is lm_sensors call for {cpu_temp} which I did not touch... mainly because we had various issues/prs wanting to fix {cpu_temp}.

• (nolink to various cpu_temp issues/prs) https://github.com/ultrabug/py3status/pull/2029.

Benchmark.

branch             case         mean_seconds      vs_master
-----------------------------------------------------------
master             default       0.000094523       baseline
master             wide          0.003934389       baseline
sysdata-update     default       0.000081974  13.28% faster
sysdata-update     wide          0.003869354   1.65% faster

[lasers]

@anarcat for potential zfs testing

[anarcat]

hey, i'm sorry but i can't test this as i don't run zfs on my desktop anymore and i don't even run py3status either. :)