Fix: gpu detection losing

[dataCenter430]

Problem

After sleep (lid close or idle), Jan showed "No GPU detected" on Linux. The GPU still worked for display/HDMI, but model inference fell back to CPU.

Cause

• Hardware/GPU info was cached once at startup (OnceLock) and never refreshed.
• NVML was initialized once; after resume the driver state was reset and the handle became stale.

Solution

Backend (Rust)

• Cache invalidation: Replaced OnceLock<SystemInfo> with RwLock<Option<SystemInfo>> so the cache can be cleared.
• New command: refresh_system_info clears the cache and (on Linux) invalidates the NVML handle so the next detection re-initializes.
• NVML: Switched to RwLock<Option<Nvml>> + with_nvml(); added invalidate_nvml() on Linux so NVML is re-inited after resume.

Frontend

• API: Added refreshHardwareInfo() to the hardware service (no-op on web, invokes refresh_system_info on Tauri).
• Auto-refresh: When the app window becomes visible again (e.g. after resume), we call refreshHardwareInfo() then re-fetch hardware and update the store (GlobalEventHandler + visibilitychange).
• Manual refresh: Settings → Hardware → GPUs card has a "Refresh" button to re-detect GPUs on demand.

Testing

• Linux (e.g. Zorin): sleep → wake → focus Jan → GPU should be detected again; or open Settings → Hardware and click Refresh.

Fixes Issues

• Closes bug: Jan loses GPU detection after system sleep on Linux (Zorin) #6447

Self Checklist

• Added relevant comments, esp in complex areas
• Updated docs (for bug fixes / features)
• Created issues for follow-up changes or refactoring needed

[dataCenter430]

hey, @louis-jan @Vanalite how are you?
are you available to review the PR?
thank you.

[louis-jan]

We will take a look

[louis-jan]

Will take a bit more time to test across platforms. ETA: tmr

[dataCenter430]

Will take a bit more time to test across platforms. ETA: tmr

okay, thanks.
pls let me know if you need anything.

[dataCenter430]

Will take a bit more time to test across platforms. ETA: tmr

Hi, @louis-jan how are you? 👋
was the test completed?
I think it should be good, since everything looks fine on my end.

[dataCenter430]

hey, are you available to review the PR?
thanks