fix: memoize static disk-name resolution so the Dashboard 2s temp poll stops re-querying WMI/SMART

[laurentiu021]

Summary

Fresh ultra-audit finding (performance, P2), verified against current source.

RefreshTemperaturesAsync calls TemperatureService.ReadAllAsync() with includeStorage defaulting to true. On an elevated session, the Dashboard's 2-second temperature poll (the Dashboard is the app's default/home tab) therefore ran, on every tick:

• GetDiskNamesFromWmi() — a Win32_DiskDrive WMI query, and
• EnrichStorageNamesAsync → DiskHealthService.CollectAsync() — an MSFT_PhysicalDisk enumeration plus a per-disk MSFT_StorageReliabilityCounter (SMART) association walk, which the code's own comments call "by far the heaviest part of a read."

All of that was spent solely to relabel LibreHardwareMonitor storage sensors with disk friendly-names — static hardware identity that never changes. (The 10s resource sampler was already optimized to skip this with includeStorage:false; the more frequent Dashboard poll was not.)

Fix

Both resolutions are memoized once, mirroring the existing _nvApiInitTried "resolve static hardware once" pattern:

• _cachedWmiDiskNames (??=, under the existing _sensorLock).
• _cachedStorageFriendlyNames behind a SemaphoreSlim gate (double-checked) so the first resolution is race-safe when the 2s poll, the user's Refresh, and the 10s sampler overlap.

After the first read the hot poll only calls LibreHardwareMonitor's Update() for live temperatures. _enrichGate is disposed in Dispose().

Regression test (integration)

EnrichStorageNames_ResolvesOnce_AndCachesFriendlyNames — drives the private enricher twice and asserts _cachedStorageFriendlyNames is populated once and reference-stable across calls (proving it resolves once). DiskHealthService.CollectAsync() degrades to an empty list on a WMI-less CI host but still caches, so the assertion is deterministic.

Checks

• All 3 projects build 0 warnings / 0 errors
• Leak scan on diff: clean
• Behavior unchanged: the same friendly-names are applied; they're just resolved once instead of every 2s.