fix(worker): report RAM alongside GPU memory

[localai-org-maint-bot]

Summary

• report system RAM during worker registration even when GPU memory is visible
• refresh available_ram on every heartbeat so GPU detection changes cannot leave a stale RAM metric
• add deterministic GPU-visible regression coverage and document unified-memory expectations

Fixes #11051

Testing

• go test -count=1 ./core/services/worker
• git diff --check HEAD^ HEAD

Documentation

Updated docs/content/features/distributed-mode.md to explain independent RAM/VRAM reporting and unified-memory behavior.

[mudler]

@localai-org-maint-bot surface this in the UI as well

[localai-org-maint-bot]

Done in 20a83b61: the Nodes roster and node detail view now show used / total system RAM alongside VRAM when the worker reports RAM capacity. Added Playwright coverage for both views; focused UI lint has 0 errors, the React production build passes, and go test -count=1 ./core/services/worker passes. Playwright could not run locally because this environment has no browser and its Chromium download is blocked with HTTP 403, so the PR run will exercise those browser assertions.