CPU reductions (same charts/features)

Cache cpuidle names (was re-reading name sysfs every tick × cores)
Cache CPU sysfs dir list + register cpufreq/throttle once
Map for per-core prev lookups (was O(n²) .find)
Dedupe meminfo/vmstat once per tick
Cache iface speed/duplex/mtu/qlen (30s)
Cache inotify limits (30s)
Gate registerChart for IRQs, cpuidle, peardock mapped charts
Remove leftover profile() stderr timers
Cache os.cpus().length in docker collector
Grow fd-cache buffer past 64KB for /proc/interrupts etc.

server/services/collectors/ebpf.js

@@ -295,7 +295,8 @@ export class EbpfCollector extends EventEmitter {
         chart: 'ebpf.vfs',
         context: 'ebpf.vfs',
         ts,
-        values: { read_pages: rate('pgpgout'), write_pages: rate('pgpgin') },
+        // pgpgin = pages read from block devices; pgpgout = pages written out
+        values: { read_pages: rate('pgpgin'), write_pages: rate('pgpgout') },
       },
       {
         chart: 'ebpf.socket',