- ui: add uptime% chip (green >=99, red below) after ssh; remove "up" word
- ui: all shadows use #0004 shadow-color; modal shadow-lg derives from it
- ui: status modal draws a datetime tick mark every 20 history entries from the newest
- server: drop max_points/downsampling/bucketing; metrics served raw, plots share a ~24h window
- docs: fix ps --no-headers in top example; document raw metrics API
- server: inject <meta name=title/description> into index.html from config title + per-group host patterns (shared with --check recap)
StatusPayload was re-resolving grouped machines from the raw
cfg.Groups configs, so glob hosts like a3-dott{1..12} showed up as a
single machine with an unresolved pattern host and no probe state.
Iterate the expanded cfg.Machines once, bucketing by group.
Adds TestStatusPayloadExpandsGlobs regression test.
Hosts accept brace groups expanded at load time: enumeration {1,2,3},
inclusive ranges {2..5,8..10} with '..' syntax, and labels {a3,a4};
several groups multiply in cartesian fashion. Leading zeros are
preserved, each expanded machine gets its host as unique ID/name.
Malformed patterns (unmatched/nested braces, non-integer or descending
ranges, extra dots) and patterns expanding past 10000 machines fail at
startup. Adds table tests plus a FuzzExpandHostPattern target; the
safety cap keeps fuzzing and real configs from blowing up memory.
aziis98