feat(ui): tab the Settings dialog by topic, widen it 80%, and turn on afterglow by default

[dc0sk]

What

Two changes DC0SK asked for together, then three refinements from on-screen feedback.

1. Spectrum afterglow (FR-PAN-14) now defaults to 500 ms, instead of
   off. It shipped off-by-default and was easy to never notice; a gentle
   trail now shows out of the box, and 0 in Settings still turns it off
   for anyone who tried it and prefers it off. A config from before the
   feature existed picks up the new default too (it never had an opinion).

2. Settings is tabbed by topic — Connection, Peers, Spotting, Audio,
   K-Pod, KPA1500, Backup — instead of one long scrolling column, and
   80% wider (500 → 900 px). It had grown crowded as features
   accumulated (KPA1500, spotting networks, afterglow, K-Pod macros, backup —
   none of it existed when it first shipped as one column).

3. Fixed (screen feedback): the dialog no longer resizes per tab. A
   first cut used max_height, which sizes the container to whichever
   tab's content is tallest/shortest — caught on screen.

4. Fixed (screen feedback): the new default wasn't applied to an existing
   config. The config already had spectrum_afterglow_ms = 0 saved
   explicitly by an earlier build — #[serde(default = "...")] only
   supplies a value for a field the file never had, not one it has. This
   needed an actual one-time migration (FR-CFG-09): Config gained
   afterglow_default_migrated: bool (false only for a file that
   predates the flag), and Config::load promotes an unmigrated 0 to
   the new default exactly once — a 0 set after upgrading is
   respected, never overridden again.

5. Settings is now a detached window (screen feedback), not a modal
   dialog. Same request as ci: fix Windows build (libopus/OpenSSL from source) #3, one layer up: DC0SK asked for the window
   itself, specifically to drop the in-content Close button, whose position
   moved with each tab's height — the same symptom as the resizing dialog.
   Settings now opens/closes via iced::window, alongside KPA1500/Networks/
   Diagnostics, at a fixed 900×720, with its own WindowClosed handling
   and ESC support, and no in-content Close button at all — the OS
   window controls and ESC replace it, matching what a real window buys you
   over a wider modal.

Design notes (mine — please overrule)

• "Wider by 180%" is ambiguous between "1.8× the current width" (900 px)
  and "an increase of 180%" (1400 px). I went with 900 px.
• The other detached windows (KPA1500, Networks) keep an in-content "Done"
  button alongside their OS/ESC close. I dropped Settings' entirely rather
  than keep one and pin its position, since removing it is what DC0SK asked
  for and directly resolves the complaint. Say if you'd rather it match the
  sibling windows' convention instead.

Tests

• fr_ui_23_every_settings_section_is_reachable_through_a_tab (structural,
  reading only the code above its own module): the window opens through
  settings_window_settings(); view/title/WindowClosed all route
  self.settings_window; opening resets to Connection; every tab has a
  button; every section from the old column is still shown by some tab; the
  widened+fixed window size (900.0 × 720.0) is present; no in-content
  Close button exists.
• fr_pan_14_afterglow_setting_persists_and_is_bounded (k4-config)
  extended for the new default; AFTERGLOW_DEFAULT_MS pinned as a literal.
• fr_cfg_09_afterglow_migrates_once_and_respects_a_later_choice
  (k4-config, new): unmigrated file, already-migrated file with a
  deliberate 0, fresh Config.
• Sabotage: the structural guard caught a deleted tab button, a
  dropped section, a reverted width, max_height reintroduced, a shrunk
  window size, and a reintroduced Close button — six distinct regressions,
  all caught, then restored. The migration test caught migrate() ignoring
  the flag.
• A slip of mine, caught before merge: the migration test's first TOML
  fixture omitted tune_step_hz (Prefs's one field with no serde
  default), so Config silently failed to parse and fell back to
  Config::default() — the test failed, but for the wrong reason. Fixed
  and re-verified it still catches the real bug.

Verified

cargo fmt --all -- --check, cargo clippy --workspace --all-targets --all-features, cargo clippy -p k4remote --features kpod --all-targets —
all clean. cargo test --workspace — all green. cargo xtask — 0 R3 gaps.

Validation parked with DC0SK

• Open Settings — does it now behave like a real window (its own
  taskbar/dock entry, moves independently of the main window), stay a
  fixed size across all seven tabs, and close cleanly via ESC and the
  window's own close control? — Confirmed 2026-09-22.
• Does 900×720 fit comfortably on your screen? — Confirmed 2026-09-22.
• Confirm the afterglow field now shows 500 (migrated) rather than 0,
  and that a live band shows the trail. — Confirmed 2026-09-22.

🤖 Generated with Claude Code