Reproducer: the launcher adds nvidia to lists that never had it, and cannot restore a list that was only nvidia

Reviewed head: 48bbdbf6022e482a1ccde831e517b56e47438c8a

Append either case to the end of test/shell.d/launch-activity-test.sh (it reuses that file's write_conf, launch and shown_gpus helpers and its stubbed omarchy-hw-nvidia-suspended), then run bash test/shell.d/launch-activity-test.sh. Each case was run on its own copy of the file.

Case 1: a list without nvidia, card awake (or no NVIDIA GPU at all)

omarchy-hw-nvidia-suspended also exits 1 when the machine has no NVIDIA display device (the PR's own detector test "a machine with no NVIDIA display device is not reported as suspended"), so the stub returning false covers both an awake card and a machine without one.

write_conf "amd intel"
OMARCHY_TEST_NVIDIA_SUSPENDED=false launch
[[ $(shown_gpus) == "amd intel" ]] || fail "a list without nvidia is left alone while the card is awake" "$(<"$btop_conf")"
pass "a list without nvidia is left alone while the card is awake"

Actual output at 48bbdbf:

ok - a suspended NVIDIA GPU is left out of shown_gpus
ok - an awake NVIDIA GPU is put back into shown_gpus
ok - a list without nvidia is left alone
ok - Activity launches without a btop config
#* Set which GPU vendors to show.
shown_gpus = "nvidia amd intel"

custom_gpu_name0 = ""
not ok - a list without nvidia is left alone while the card is awake

Exit status 1.

Case 2: a list that is only nvidia

write_conf "nvidia"
OMARCHY_TEST_NVIDIA_SUSPENDED=true launch
OMARCHY_TEST_NVIDIA_SUSPENDED=false launch
[[ $(shown_gpus) == "nvidia" ]] || fail "a list of only nvidia is put back once the card is awake" "$(<"$btop_conf")"
pass "a list of only nvidia is put back once the card is awake"

Actual output at 48bbdbf:

ok - a suspended NVIDIA GPU is left out of shown_gpus
ok - an awake NVIDIA GPU is put back into shown_gpus
ok - a list without nvidia is left alone
ok - Activity launches without a btop config
#* Set which GPU vendors to show.
shown_gpus = ""

custom_gpu_name0 = ""
not ok - a list of only nvidia is put back once the card is awake

Exit status 1. The first launch writes shown_gpus = ""; the second skips the file because of the [[ -n $shown ]] guard, so nvidia never comes back.

Symlinked btop.conf is replaced by a regular file

sed -i without --follow-symlinks replaces a symlinked ~/.config/btop/btop.conf with a regular file on the first launch while the card sleeps, detaching it from the user's dotfiles.

Suggested change: Use sed -i --follow-symlinks, as omarchy-theme-set-vscode does.

write_conf "nvidia amd intel"
mkdir -p "$tmp_dir/dotfiles"
mv "$btop_conf" "$tmp_dir/dotfiles/btop.conf"
ln -s "$tmp_dir/dotfiles/btop.conf" "$btop_conf"
OMARCHY_TEST_NVIDIA_SUSPENDED=true launch
[[ -L $btop_conf ]] || fail "a symlinked btop.conf stays a symlink" "$(ls -l "$btop_conf" | cut -c1)"
[[ $(shown_gpus) == "amd intel" ]] || fail "a symlinked btop.conf is edited through the link" "$(<"$btop_conf")"
pass "a symlinked btop.conf is edited through the link"

ok - a suspended NVIDIA GPU is left out of shown_gpus
ok - an awake NVIDIA GPU is put back into shown_gpus
ok - a list without nvidia is left alone
ok - Activity launches without a btop config
-
not ok - a symlinked btop.conf stays a symlink

Verified:

• btop 1.4.7 calls Gpu::Nvml::init() at startup whenever shown_gpus contains nvidia, whatever boxes are shown (src/linux/btop_collect.cpp:371-374).
• The detector reads only cached sysfs fields (vendor, class, power/runtime_status), and returns suspended only when every NVIDIA display-class function is suspended; transitional and unsupported states keep nvidia in the list.
• The shipped list nvidia amd intel becomes amd intel while the card is suspended and is restored byte for byte once it is awake.
• Super+Ctrl+T resolves to omarchy-launch-activity and keeps the org.omarchy.btop app id and float rule.
• Both new test files pass at the head and fail without the scripts.

Rebase and overlap: the PR conflicts with current quattro in default/hypr/bindings/utilities.lua (neighbouring bindings changed; the Activity line itself did not). Open PR #8661 changes the same Activity line to { tui = "btop", focus = "current-workspace" }, so whichever lands second needs to carry the other's behaviour.

Optional notes:

• The edit persists: after an Activity launch while the card sleeps, a btop started from a terminal omits NVIDIA until Activity is next opened with the card awake.
• Reverting the binding to { tui = "btop" } survives the tests; a one-line grep -F check like the one in launch-1password-test.sh would cover it.