Add an Arm benchmark data point; name Arm CPUs

[jacotay7]

This adds the device benchmark from cfl-test-bench: an Ampere Neoverse-N1 (aarch64), pinned to 16 cores, with an RTX 4060 and an RTX A400, running getframes 2.4.0.

• Artifacts: benchmarks/device-results-neoverse-n1-rtx4060.json (CPU and GPU) and ...-rtxa400.json (GPU).
• Table: benchmarks/device-results-neoverse-n1.md puts both cards and the N1 CPU next to the existing Ryzen 9 9950X3D and RTX 5090 numbers. docs/guides/gpu.md links to it.
• Fix: bench_devices.py reported Arm CPUs as aarch64. It now reads the model from lscpu (Neoverse-N1) when /proc/cpuinfo has none. GPU naming was already right: it uses CuPy's current device.
• Findings:
  • The GPU's advantage grows with the detector: 1.2× at 80×80, 18× at 1024².
  • The RTX 4060 runs 1.1× to 5.6× the A400, the gap growing with detector size.
  • 16 N1 cores reach a steady 0.40–0.44× the 16-core Ryzen.
• Tests on the bench: the full suite and the GPU tests pass.

🤖 Generated with Claude Code