v1.4.13

Settings the user changes on the headset now survive a power cycle, the Arctis 7
ChatMix dial is read correctly, and the update banner stops lying about what is
available when a distro repository lags behind GitHub.

Fixed

• A headset that powers off and on gets its settings back. Sidetone, mic
  volume, the volume limiter and the on-device EQ live in the headset, not the
  dongle. Switch it off and they are gone; on wireless families the dongle never
  leaves the port, so a power cycle fires no USB event and ASM never re-applies
  anything. The user's saved settings stayed on the Devices page while the
  hardware ran on firmware defaults — reported for the Nova 3 Wireless in #221,
  but the gap covers the whole wireless line. GG re-pushes on the connection
  event; this is the same re-push. The connection status already drives the
  audio redirect in on_device_status_changed; it now also schedules a settings
  replay from the profile's own device_init frames — mode switches included, so
  the two paths cannot drift — plus any explicit setting device_init does not
  cover and the stored EQ curve. Runs off the listen thread, debounced on the
  last push. (#221)
• The Arctis 7 ChatMix dial no longer freezes at the position it had when the
  daemon started. SteelSeries' spec declares game_chat_status as out
  [0x06, 0x24], back [0x06, 0x24, game, chat] — it is not part of the
  battery reply, and this dongle pushes nothing when the dial turns. The profile
  asked for it exactly once in device_init, and the status poll only asked for
  battery, so the mix ASM showed stayed pinned for the whole session.
  manage_mix_change applies those values to the Game and Chat channel volumes,
  which meant a dial at one end at startup pinned a channel there forever. The
  poll now sends status.extra_requests after the main request, each reply
  mapped by its own starts_with, and a failing extra never costs the battery
  reading that already landed. (#220)
• The Arctis 7 2019 ChatMix dial reported the wrong volume.
  game_chat_status's game/chat fields are a signed byte of attenuation from
  0x00 (full volume), counting down through 0xFF…0xC0 toward mute — the same
  convention hardware_eq.py already decodes for EQ gain. The plain unsigned
  percentage(0, 100) mapping read the loudest position as 0% and clamped every
  attenuated one to 100%. A signed percentage parser is added and pointed at
  media_mix and chat_mix in arctis_7.yaml. Based on reporter jsuvanto's
  raw USB capture on #220 — SteelSeries' own spec only declares game/chat as
  plain uint8 with no documented scale, so real hardware behaviour is the only
  source here.
• Sonar EQ live-apply no longer lies about success on restricted systems.
  On SteamOS and other setups that start clients restricted (#181), pw-cli set-param can be refused by PipeWire with "Operation not permitted" the same
  way pw-link is. Unlike the link-permission fix, this call site had no
  repair-and-retry, so the refusal was silently eaten and _ApplyWorker
  reported the apply as successful anyway — the conf on disk was correct but the
  running graph kept the old values, matching the "presets do nothing" symptom.
  grant_props_permissions() is added, mirroring grant_link_permissions()'s
  already-confirmed repair (raise the owning client to rwxml, retry once), and
  wired into set_filter_controls. _ApplyWorker now only reports success when
  the push genuinely cannot be attempted (node not up yet).
• The udev dialog no longer writes rules into the container on distrobox
  installs. udev_checker already learned to read the host's rules through
  distrobox-host-exec, but write_udev_rules and reload_udev_rules still
  targeted the container's own /etc, where udev never looks — the dialog's
  "Run" button reported success and changed nothing. Rules are now staged under
  ~/.cache (shared with the host via distrobox's bind mount) and installed
  through distrobox-host-exec, trying sudo then pkexec. container.py
  centralizes running_in_container() / host_exec() / host_distro() /
  host_is_immutable(), replacing three drifted copies. On an immutable host
  reached through a container the dialog now offers the matching
  scripts/distrobox/*.sh installer instead of fabricating an untested
  rpm-ostree command. Reported on Bazzite alongside #140.
• The "positional fallback" log line no longer drowns out real device
  reports. It fired at INFO before the "does this link already exist" check,
  so it went out on every watchdog pass whether or not anything was created — on
  a Nova Pro that is every loopback, since the headset's sink names its ports
  AUX0/AUX1. Three to four INFO lines every five seconds run about 270
  characters each, filling the "Recent daemon logs" section of a bug report with
  nothing but noise. It now logs INFO the first time for a given pair and DEBUG
  after.
• The update banner no longer loops when a distro repository lags behind
  GitHub. COPR lagged a release by roughly seven hours in discussion #140: the
  banner announced 1.4.12 while the user's dnf could still install only 1.4.11,
  so "Update now" ran an upgrade that found nothing and the banner came back.
  The same pattern would have hit APT and pacman installs. The user's own
  repository is what the upgrade command runs through — it is the only answer
  that is actionable — so it now decides first. GitHub is the fallback for pipx,
  pip and AUR installs.

Added

• A probe for the Arctis 7 ChatMix dial in scripts/reverse-engineering/,
  with a README explaining how to capture and analyse the raw USB report. Used
  to confirm the signed-attenuation decoding in #220.