v4.8.0

🧑‍💻 AI runtimes run by another account. When Ollama, oMLX or another runtime runs
under a separate user — an ai account serving models for you — SiliconScope now
detects it and asks its API, as it does for your own. Python runtimes started from a
venv on Homebrew's or python.org's Python (mlx_lm, vLLM, MTPLX …) and oMLX installed
with pip are detected too, under any account. Other users' processes are used for
detection only; the Processes card still lists yours.
(#85, @martin-kufner)

🎮 GPU processes in Fleet. On a Linux GPU box, each process now shows its share of
the card (SM %), and graphics processes — a display server, a desktop shell — are
listed with a label. Needs Linux agent 1.5.3.

Fixes

• The dashboard and menu bar could stop updating until the app was restarted, when a
  Bluetooth query for AirPods battery hung. That query now runs in the background with
  a time limit, and any part of sampling that stalls is logged by name.
  (#86, @Borda)
• On chips with no ANE state channel (M4 Pro, M2 Pro), the ANE row shows the engine's
  memory traffic instead of watts alone.
  (#87, @ebekebe)
• sscope-cli --power-debug converts each channel by its own unit: an M6's 0.014 W
  GPU no longer prints as 13,733 W.
  (#84)
• Linux agent: LM Studio's log stream no longer keeps running after the agent is killed.

Agents: Linux 1.5.3, Mac 1.4.2 (other accounts' runtimes, and the time-limited
tools above).