Add RX audio stream diagnostics

[rfoust]

Summary

• Add RX audio stream diagnostics for feed rate, slow delivery, late arrivals, packet gaps, stream format, and packet freshness.
• Surface user-friendly audio health in Network Diagnostics with per-stream rows and trend charts.
• Combine the Audio and Audio Feed views into a single Audio tab.
• Reset audio diagnostics when the remote audio stream stops and restarts.
• Restyle Network Diagnostics to match the AetherModem window, including rounded panels, compact Details layout, and updated graph/table/log styling.

Verification

• git diff --check
• cmake --build build --parallel 8
• Checked clean merge compatibility with PR 2879 in both merge orders.

[Copilot]

Pull request overview

This PR adds per-RX-audio-stream diagnostics (rate/deficit/late/gaps/freshness/format) sourced from PanadapterStream, resets those diagnostics when the RX audio stream lifecycle changes, and surfaces the results in the Network Diagnostics UI (new Audio tab table + graphs) with an updated dialog style.

Changes:

• Add PanadapterStream::AudioStreamDiagnostics tracking, snapshot, and reset hooks for VITA audio packets.
• Expose audio diagnostics through RadioModel, and reset diagnostics on RX audio create/adopt/remove events.
• Expand/restyle NetworkDiagnosticsDialog to display audio stream health, per-stream rows, and timing/feed trend charts.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/models/RadioModel.h	Exposes audio stream diagnostics getters/reset on the model.
src/models/RadioModel.cpp	Implements diagnostics getter/reset and wires resets into RX audio stream lifecycle.
src/gui/NetworkDiagnosticsDialog.h	Extends sampling/state to include RX audio feed/late/gap/format fields and new UI widgets.
src/gui/NetworkDiagnosticsDialog.cpp	Implements new RX audio formatting/health logic, table + graph UI, and dialog restyling.
src/core/PanadapterStream.h	Defines audio diagnostics/tracker structures and adds snapshot/reset APIs.
src/core/PanadapterStream.cpp	Records per-audio-stream timing metrics from incoming packets and provides snapshot/reset behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rfoust]

Addressed in the amended commit: kAudioSampleRate now references AudioEngine::DEFAULT_SAMPLE_RATE instead of duplicating 24000.

[rfoust]

Addressed in the amended commit: removed the duplicate manual audio gap/jitter resets from start, rebindToEphemeralPort, startWan, and stop; resetAudioStreamStats() is now the single reset path.

[rfoust]

Addressed in the amended commit: formatAudioHealth() now returns an AudioHealthStatus with an AudioHealthState enum plus display text, and the label color is selected from the enum instead of comparing user-facing strings.

[aethersdr-agent]

Thanks @rfoust — nice diagnostic work, and the per-stream feed-rate / deficit / late-arrival metrics will be genuinely useful when triaging RX audio glitches in the field. The wiring through recordAudioStreamPacketLocked inside the existing stats-mutex region looks correct, and resetting on every (re)start / stop / clearRegisteredStreams is the right call.

A few comments — none are blockers.

Copilot findings — all three look valid:

1. kAudioSampleRate = 24000 in PanadapterStream.cpp:24 duplicates AudioEngine::DEFAULT_SAMPLE_RATE (defined at src/core/AudioEngine.h:71, used in ~40 places across AudioEngine.cpp). Worth referencing the shared constant so the deficit/feed-rate math stays correct if the engine sample rate ever changes. A simple #include "AudioEngine.h" and constexpr int kAudioSampleRate = AudioEngine::DEFAULT_SAMPLE_RATE; would do it.

2. Duplicate field resets in start / rebindToEphemeralPort / startWan / stop. Each call site now does:

   m_audioPacketTimerStarted = false;
   m_previousAudioPacketGapMs = 0;
   m_audioPacketJitterEstimateMs = 0.0;
   resetAudioStreamStats();  // ← also resets all three

   resetAudioStreamStats() already covers those three members. Drop the manual lines at the four call sites and let resetAudioStreamStats() be the single source of truth — that's exactly what it's there for.

3. Brittle string comparisons in the audio-health styling (NetworkDiagnosticsDialog.cpp:2099). The if (audioHealth == "Healthy") ... else if (audioHealth == "Audio stream is arriving too slowly" ...) branches will silently fall to the warning-color fallback if formatAudioHealth()'s wording ever drifts. Acknowledging this mirrors the existing networkQuality() pattern in the same function, so it's not out of step with the file. Optional follow-up: have formatAudioHealth return an enum (or a {status, displayText} pair) and switch on the enum. Not blocking.

Scope note (minor): The PR description does call out the restyle, so it's not hidden, but the kNetworkDiagnosticsStyle block + Audio/Audio-Feed tab consolidation are really a separate UI refresh from the diagnostics feature. Easier to bisect later if they land as two commits, but it's a one-author PR so I won't push on it.

Things I didn't find issues with:

• The Locked suffix convention is honored — recordAudioStreamPacketLocked is called from inside the m_statsMutex-held region in processDatagram, and audioStreamDiagnostics() / resetAudioStreamStats() both re-acquire the lock for snapshot/reset. Looks thread-safe.
• audioPayloadFrames correctly returns 0 for unknown PCCs and recordAudioStreamPacketLocked early-exits on frames <= 0, so unknown codecs don't pollute the per-stream tracker.
• Reset on clearRegisteredStreams matches the existing pattern from #2738.

[jensenpat]

Slice 0 RX flow: untouched — only adds passive observation in the existing audio-decode branch. Low risk.
Transmit interlock, command ordering, timers: unchanged.
GUI restyle: affects only NetworkDiagnosticsDialog. Per CLAUDE.md, visual design is aligned per Jeremy's chat feedback.

[ten9876]

Claude here — merged. Thanks @rfoust, RX audio diagnostics is
exactly the kind of observability the project has been missing.

What stood out on review:

The thread-safety design is intentional and well-marked. Naming
`recordAudioStreamPacketLocked()` explicitly to flag that it's
called with the mutex held, making `resetAudioStreamDiagnostics()`
a `Q_INVOKABLE` so RadioModel can dispatch it cross-thread via
`QMetaObject::invokeMethod(..., Qt::AutoConnection)` — these
patterns disappear into bugs when they're not done deliberately.

The health-status rules are conservative in the right way. Compound
late-packet condition (`latePacketsPerSecond >= 2.0` OR `>= 1.0` AND
`feedDeficitMs >= 10.0`) catches recurring problems while ignoring
the inevitable single-packet jitter at the noise floor. The 95%
feed-rate threshold likewise allows for small dropouts without
crying wolf.

And the lifecycle-reset discipline — 5 reset sites covering every
RadioModel path that creates/adopts/removes a stream — prevents the
classic "stale counter from the previous stream session" footgun.

Auto-merge clean against #2879's TCI tab work despite both touching
`NetworkDiagnosticsDialog` — you verified both merge orders before
submitting, which is the right way to ship parallel UI work.

Ships in v26.5.3. Operators chasing audio-stream weirdness now have
real visibility into per-stream feed rate, late packets, and clock
skew rather than guessing from the audio buffer dial.

73,
Jeremy KK7GWY & Claude (AI dev partner)