v5.16.0

Added

• Agent (AMD GPU): new gpu_amd_smi sampler collects AMD GPU
  utilization, memory, power, temperature, and clock metrics via ROCm
  SMI, with a matching viewer dashboard. NVIDIA GPU metrics now carry a
  vendor="nvidia" label so multi-vendor fleets can be distinguished.
  (#973)
• Agent (AMD GPU): additional gpu_amd_pmu sampler reads AMD GPU
  hardware performance counters directly. It has high CPU and memory
  overhead and is therefore disabled by default; enable it explicitly
  and set the desired perf_level. (#973)
• Agent (drive health): drivehealth reports per-drive temperature via
  SMART pass-through ioctls, alongside the existing health metrics.
  (#992)
• Agent: sampler health classification and a new rezolus agent status
  command that reports each sampler as active/disabled/failed. (#977)
• Recorder: perf-record-style command wrapping — rezolus record can
  launch and record around a target command, capturing metrics for the
  span of that process. (#981)
• Viewer: "Save as Report" in two-file A/B compare mode, exporting the
  current comparison as a shareable report. (#960)
• Viewer: support for arbitrary "simple capture" parquets, with
  per-source detection and a Metrics browser for exploring their
  columns. (#989)

Fixed

• Viewer: live mode now refreshes metadata so the chart query window
  tracks the TSDB as it grows. (#974)
• BPF: samplers now emit a warning when a program fails to attach due to
  a missing kernel symbol, instead of failing silently. (#972)

Changed

• Every subcommand's --help output was reworked to be self-explanatory
  and agent-usable, and the README was reorganized to emphasize use
  cases and the full feature set. (#976, #987, #983)