AetherSDR v26.8.1 — Hermes-Lite 2, SQLite Settings & Capability-Gated UI

118 commits since v26.7.4.1.

The Hermes-Lite 2 grows from an experimental backend into a radio you can
operate: four independent receivers, the SSB voice chain, CW and RTTY
decoding, AX.25 packet with a mailbox proven on the air, band switching with
hardware filters and preamp, host-side memory channels, per-MAC operating
state restore, and a Radio Health dialog. FlexRadio remains the supported
target.

Client settings move to SQLite (RFC aethersdr#4603) with transactional saves,
integrity checks, verified backups, credentials in the OS keychain, a
--config command line and a Settings Browser. RadioCapabilities now gates the
UI, so every Flex-only surface hides itself on a backend that has no such
thing.

Qt 6.8.3 LTS is the floor and the pin everywhere — the CI image, both
AppImage architectures, the Windows installer and both macOS legs. The Linux
AppImage runs natively on Wayland; the aarch64 AppImage gets GPU spectrum
rendering.

The Intel Mac DMG trades speech-to-text for reach: it declares and honours a
macOS 12.0 floor, because the speech runtime is published at a macOS 15.5
floor and one library's floor becomes the whole bundle's. Apple Silicon keeps
speech-to-text.

TCI PTT now keys the slice the client asked for.

AetherSDR v26.7.4.1 — TCI rig control hotfix

One fix on top of v26.7.4: TCI vfo: SET now confirms the frequency the
radio actually reached, restoring WSJT-X rig control and fixing relative
tuning from control surfaces (aethersdr#4500, aethersdr#4493).

AetherSDR v26.7.4 — Copy Assist, NR2 & Demo Mode

Pre-fill issue reports with a redacted log tail. Principle VII.

Fixes aethersdr#3705

## Summary

- Pre-fill Help → Support → Submit Bug Report with system/radio context and a bounded recent-log tail.
- Re-run centralized privacy redaction at the public issue-render boundary and fall back to the clipboard when the encoded GitHub URL would be too long.
- Keep SmartLink account names and GPS/location coordinates out of logs at their source, including the WAN certificate-decision path.
- Add centralized name/coordinate redaction as a fail-safe so those fields cannot survive into disk logs or the public issue body if a future logging path submits them.
- Replace the absolute no-personal-data claim with a concrete privacy notice and a prompt to review before submission.

## Verification

- Full macOS RelWithDebInfo build with 8 parallel jobs.
- `async_log_writer_test` passes with planted SmartLink names and GPS/location coordinates absent from disk logs.
- `issue_report_test` passes with the same planted fields absent from the generated public report.
- `tools/check_engine_boundary.py --strict` passes with no new violations.

Squashed-from: aethersdr#4314

Co-authored-by: aethersdr-agent[bot] <273844287+aethersdr-agent[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Pat Jensen <patjensen@gmail.com>

ci: auto-cancel superseded workflow runs via concurrency groups (aeth…

…ersdr#4346)

## Problem

The four PR check workflows — **CI**, **CodeQL**, **Accessibility
Check**, **Engine Boundary Check** — had no `concurrency` group. Every
push to a branch stacked a fresh full run on top of the still-queued
older ones, so a rapid commit stream built up dozens of superseded runs
and starved the runner pool. A ~35-run backlog (mostly
`feat/asr-phase1-build`) just had to be cancelled by hand.

## Fix

Add a concurrency group keyed on `workflow + ref` to each of the four
workflows:

```yaml
concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
```

A newer commit on the same PR/branch now auto-cancels the in-progress
older run.

**The `main` guard is deliberate.** `cancel-in-progress` evaluates to
`false` on `refs/heads/main`, so:
- ci.yml / codeql.yml **post-merge `push: main` guard** always
completes,
- CodeQL's **weekly scheduled scan** (runs on the default branch) always
completes.

Only PR-branch runs (`refs/pull/N/merge`) get cancelled when superseded.

## Scope

- Touches only the four PR-check workflows.
- Release/artifact workflows (`appimage`, `macos-dmg`,
`windows-installer`, `streamdeck-plugins`) are **left untouched** — they
weren't part of the backlog and shouldn't cancel mid-build.
- `docker-ci-image.yml` and `mirror-aether-gate.yml` already have their
own `concurrency` blocks.

## Validation

- All four files parse as valid YAML; concurrency schema asserted
(`group` + `cancel-in-progress` present).
- No job logic changed — pure top-level `concurrency:` addition (28
insertions, 0 deletions).

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

fix(waterfall): unify pane height split (aethersdr#4322) Principle VII.

Derive the waterfall pane height as the remainder of the shared spectrum split,
then use that value for image allocation, painting, and divider-drag
reallocation.

This prevents the source image and destination rectangle from differing by one
row at odd content heights and fractional split ratios, avoiding Qt stretching
or duplicating a waterfall row.

Validation included a clean RelWithDebInfo build, targeted renderer and layout
tests, the strict engine-boundary check, and all required GitHub checks.

The separate faint Metal-path artifact documented in review remains outside the
scope of this arithmetic fix.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

AetherSDR v26.7.3 — Cross-Needle Meter & Operating Polish

New meter choices, tighter radio-authoritative state, and a broad operator-workflow polish pass. 25 commits since v26.7.2. Cut by Pat KI6BCJ.

AetherSDR v26.7.2

aetherd IRadioBackend seam, MCP server, BNR on RTX 50-series/Blackwell,
per-receiver KiwiSDR passwords, searchable Radio Setup & Network Diagnostics,
local D-STAR via ThumbDV, adaptive ESSB RX filter, WAVE showcase visualizations.
123 commits since v26.7.1. See CHANGELOG.md.

AetherSDR v26.7.1

3D stacked-trace spectrum + in-process NVIDIA BNR + TX meter readouts + 60 fps panadapters.

Headlined by a new 3D stacked-trace spectrum panadapter render mode, the NVIDIA
in-process AI noise-removal engine (BNR, no container), mouse-over numeric
readouts on the TX meters, and a 60 fps per-pixel GPU panadapter path — plus
platform-based model capability gating from FlexLib and a broad wave of audio,
spectrum, and UI fixes. 29 commits since v26.6.5.

See CHANGELOG.md for the full list.

fix(gui): DX spot tooltip blocked by passband hit-test (aethersdr#3887)…

… (aethersdr#3903)

## Summary
- Hovering over a DX spot inside the active slice passband showed no
tooltip because `sliceCursorShapeAt` (passband body → `OpenHandCursor`)
set `foundCursor = true` before the spot rect loop had a chance to run.
- Fix: move `m_spotClickRects` check ahead of `sliceCursorShapeAt` in
  `mouseMoveEvent`. 7-line block reorder, zero logic changes.
- Mirrors the existing press path (`spectrumDefaultsToCrosshairAt`,
  line 5578) which already returns `false` early for spot rects —
  click and hover now behave consistently.

## Test plan
- [ ] Enable SpotHub; tune a spot fully inside the active slice passband
— tooltip appears on hover
- [ ] Same spot moved outside the passband — tooltip still works
- [ ] Filter edge resize zones (`kFilterEdgeGrabPx` grab strips) still
respond correctly
- [ ] Spot cluster indicators inside passband show `PointingHandCursor`
- [ ] Tested: macOS 15, FLEX-6600, v26.6.5 build

Fixes aethersdr#3887
Closes aethersdr#3901