[jamesbraid]

audio status is now one machine-level monitor command that lists every sound card the machine has (HAL2, MACE audio, RAD1, the Tezro's ICE1724) in one layout, or says No audio device.. It replaces hal2 status, ice1724 status and MACE audio's own audio status.

Per output, each card shows the frames moved on its own clock and the guest rate, and, only when some occurred, the frames its clock made due that it skipped, by cause:

audio: MACE audio
  analog out 1: 128 moved, guest rate 8000 Hz
    skipped: stale 14000, dry 1872, ring full 0
  host: 0 underruns, 0 frames skipped, 0 dropped, queued 19.8 ms, delay 20.0 ms, adjust +0.001%
  host output: <device> at 48000 Hz, 2 channels

• stale: the card fell behind real time and drops the lost time (HAL2 pacer backlog limit, MACE audio's quarter-second lag, ICE1724 clock jump and steering backlog, RAD1 ring laps it never walks).
• dry: the guest's ring had nothing (MACE audio empty ring, HAL2 channel reads that returned nothing, RAD1 chain end or DMA failure, dropped once and taken into the channel's own count).
• ring full: MACE audio's input ring.

The host side is not counted again. It is iris_audio::Stats for the card's outputs: underruns, frames skipped to hold latency, frames dropped on a full ring, queued sound, delay, clock adjustment, device and format. For MACE audio's two DAC voices the host line sums underruns, skipped and dropped, shows the most queued, and takes delay, adjustment and device from the first.

Wiring. New Device::audio_report (default None); the monitor prints every registered card's report. HAL2 reports through the HPC3, MACE audio through the O2 bus, and the Octane and Onyx2 now register RAD1. audio codec stays on MACE audio.

Size: 26 files, +505/-255. Production is not a net deletion: the machine-level dispatch and the skip causes cost more than the status dumps they replace.

Removed: hal2 status with its ISR, DMA, BRES, codec and AES dumps; Hal2State.iar_log and its recording; CodecAState.frames/calls/armed_ch/dry_reads; ice1724 status (converter write counts and unexpected() stay in Stats for snapshots and tests but are no longer shown); MACE audio's status text. Scripts and a clanker note that said hal2 status now say audio status.

Scrutinize

• "moved" differs by card: MACE audio's DAC moves a frame per clock tick the ring supplied (an empty ring is dry); RAD1's moved no longer includes laps (stale); the ICE1724's is its saved engine count (snapshots and 26 test reads use it), which the report copies into the meter at command time, so there is no second counter. The ICE1724 counts only stale; its failed transfers still count as moved.
• HAL2 counts codec A only; codec B and the AES channels are not counted. With no host output it still reads the channel without converting it.
• help lists audio once, and only where a card can answer.
• ICE1724 counts are as of the last timer pass.
• Counters are relaxed atomics, once per timer pass or DMA chunk.
• Ip27Machine::monitor() is a #[cfg(test)] accessor like the existing bridge().

Verified (all on the final head)

• cargo check --workspace --all-targets, default and --features jitv2.
• cargo test -p iris --lib -- hal2 mace_audio rad1 ice1724 host_audio host_pcm monitor hpc3 ip32:: ip30_machine ip53 ip27_deskside: 337 pass on Linux.
• Tests that carry the change: exact report text; HAL2's pacer stall (injected time), dry reads, no-host read and report while codec A plays; MACE audio's stall and empty ring; RAD1's laps and chain end; the ICE1724 clock jump and report; the monitor with two cards, with none, and help listing audio once and only with a card; RAD1 reachable through the Octane and Onyx2 monitors, and "No audio device." on an Origin 2000. Fifteen of them ran 50 times alone and 50 times under four CPU hogs on two pinned CPUs: no failure.
• Mutations that fail their test: HAL2 dry count removed, Octane RAD1 registration removed, monitor filter_map to find_map, HAL2 no-host conversion forced on, help duplicate-check removed.

Not run: audio status against a booted guest through the telnet monitor, a real sound device, macOS or Windows. No machine has two cards, so the several-cards listing uses two stub devices. The monitor no longer holds its device lock across the socket write; that is covered by review only (a regression test needs a stalled client).