…2784)

## Summary

Implements [the plan from
#2761](#2761). Replaces
incomplete agent attempt #2778 (which committed the API surface without
the implementation wiring).

When a slice tunes near a pan edge, the existing **Pan Follows VFO**
machinery slides the pan center so the slice stays visible. But the VFO
flag panel hanging off the slice marker is fixed-width — the slice can
be well inside the trigger boundary while the flag panel is already
clipping the pan edge. Split pairs (LockLeft + LockRight on the same
marker, #2663 / #2744) are particularly affected: PR #2744 intentionally
accepted that the outward-facing flag clips and documented \"the user
pans toward center to read it\" as the workaround.

This change removes that workaround by treating the **flag's outer
edge** as the trigger reference, not the slice frequency.

## What changes

- \`revealFrequencyIfNeeded\` gains \`leftFlagEdgeOffsetMhz\` /
\`rightFlagEdgeOffsetMhz\` defaulted to \`0.0\`. Only consumed on the
\`IncrementalTune\` path — \`CommandedTargetCenter\` /
\`RevealOffscreen\` ignore them.
- Offsets are clamped to 95% of \`triggerDistanceFromCenter\` so tiny
panadapters where flag-width > trigger-margin don't oscillate the pan
around the slice.
- \`panFollowVfo\` now computes flag-width-in-MHz from the bound
\`VfoWidget\` (\`width() * (bandwidthMhz / specPixelW)\`) and threads it
through:
- **Split pair** (\`SpectrumWidget::sliceHasSplitPartner\` returns
true): both offsets non-zero — extends the trigger on both sides since
LockLeft + LockRight flags render on both sides simultaneously.
- **Single-flag slice**: only the side where the flag currently renders
(per \`VfoWidget::onLeft()\`).
- **Collapsed / compact-mode flag**: both offsets stay 0.0 — compact
mode doesn't render a wide flag.

## Files

| File | Change |
|---|---|
| \`src/gui/MainWindow.h\` | \`revealFrequencyIfNeeded\` signature +
comment |
| \`src/gui/MainWindow.cpp\` | Implementation: clamp + apply offsets in
the trigger comparison; \`panFollowVfo\` computes offsets from VfoWidget
state |
| \`src/gui/SpectrumWidget.h\` | \`sliceHasSplitPartner(int)\` accessor
|
| \`src/gui/SpectrumWidget.cpp\` | \`sliceHasSplitPartner\`
implementation (scans \`m_sliceOverlays\` for the slice + checks
\`splitPartnerId\`) |
| \`src/gui/VfoWidget.h\` | \`onLeft()\` accessor exposing the existing
\`m_lastOnLeft\` state set by \`updatePosition\` |

Net: **85 insertions / 9 deletions** across 5 files.

## Local verification

- [x] Build clean (\`cmake --build build --target AetherSDR\`) — 401/401
link, only pre-existing warnings
- [x] Test-target link audit: none of the touched headers are linked by
isolated test targets — the new \`onLeft()\` accessor and
\`sliceHasSplitPartner()\` accessor are GUI-only and only consumed from
MainWindow.cpp. No CMakeLists changes needed.

## Test plan (manual, post-merge)

- [ ] Tune a slice toward a pan edge with a wide VFO flag visible — pan
should slide *before* the flag clips
- [ ] In split mode (LockLeft + LockRight on the same marker) — pan
should slide before *either* flag clips
- [ ] Non-flag callers (Memory recall, click-to-tune absolute jump) — no
behavior change (CommandedTargetCenter ignores the offsets)
- [ ] Compact-mode (collapsed) flag — falls back to original
slice-frequency comparison
- [ ] Toggle \`View → Pan Follows VFO\` off — pan does not slide at all
(existing override preserved)

## Closes

- Closes #2761
- Closes-as-replacement for #2778 (incomplete recovery)

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

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>