Redesign Display Settings, Wallpapers, Booru; add System Monitor, Weather

Display Settings: drag-to-arrange multi-monitor canvas with edge
snapping (locked when only one screen is connected), 8/10-bit depth,
color management presets (sRGB/Wide/EDID/HDR+EDID), SDR
brightness/saturation, ICC profile path, VRR toggle. display-set now
takes a single pre-built descriptor instead of a fixed positional arg
list, since Hyprland's monitor keyword grew well past that.

Wallpapers: fixed the current-wallpaper highlight never loading on
open, added real first-frame thumbnails for animated wallpapers
(wallpaper-thumbs + ffmpeg), bigger 2-column grid, filename filter.

Booru: separate downloads folder from Wallpapers so browsing doesn't
clutter the wallpaper picker, bigger grid, Safe/Moderate/Explicit
rating tiers replacing the old binary toggle, sort order (Newest/
Score/Random). Fixed pagination being completely broken (a Flow
anchor violation was silently disabling the whole layout) and
Danbooru's Random sort (order:random tag times out server-side --
switched to their random=true endpoint for that case).

Wi-Fi panel: added a "Scanning.../No networks found" state and a
manual rescan button -- previously showed nothing either way, making
"still scanning" indistinguishable from a real failure.

Bluetooth: auto-scans on open like Wi-Fi now, instead of requiring a
manual click every time.

New: System Monitor panel (CPU/RAM/GPU/VRAM/disk usage, NVIDIA or AMD
detected automatically) and Weather panel (IP-geolocated or a typed
city, current + hourly forecast, Open-Meteo, no API key) with a
compact taskbar badge.

modules/scripts/system-telemetry.sh

@@ -0,0 +1,62 @@
+#!/usr/bin/env bash
+# Single-shot system telemetry snapshot — CPU/RAM/GPU/VRAM/disk usage — as
+# one JSON object, so the Quickshell System panel can poll this on a Timer
+# instead of spawning half a dozen separate processes every tick.
+set -euo pipefail
+
+# --- CPU ---
+# /proc/stat's first line ("cpu ...") is cumulative jiffies since boot —
+# computing instantaneous usage needs two samples across a short gap, so
+# both reads happen inside this one invocation rather than needing
+# cross-poll state kept on the QML side.
+read_cpu() {
+    read -r _ user nice system idle iowait irq softirq steal _ < /proc/stat
+    echo "$((user + nice + system + irq + softirq + steal)) $((user + nice + system + idle + iowait + irq + softirq + steal))"
+}
+read -r cpu_busy1 cpu_total1 <<< "$(read_cpu)"
+sleep 0.2
+read -r cpu_busy2 cpu_total2 <<< "$(read_cpu)"
+cpu_busy_delta=$((cpu_busy2 - cpu_busy1))
+cpu_total_delta=$((cpu_total2 - cpu_total1))
+cpu_pct=0
+[ "$cpu_total_delta" -gt 0 ] && cpu_pct=$((100 * cpu_busy_delta / cpu_total_delta))
+
+# --- RAM ---
+mem_total_kb=$(awk '/^MemTotal:/{print $2}' /proc/meminfo)
+mem_avail_kb=$(awk '/^MemAvailable:/{print $2}' /proc/meminfo)
+mem_used_kb=$((mem_total_kb - mem_avail_kb))
+
+# --- GPU / VRAM ---
+# NVIDIA checked first via nvidia-smi (works regardless of which DRM card
+# index the driver landed on); AMD falls back to the card0 sysfs counters
+# directly — whichever's actually present decides which branch fires, so
+# this works unmodified on either machine.
+gpu_pct=-1
+vram_used_mb=-1
+vram_total_mb=-1
+if command -v nvidia-smi > /dev/null 2>&1 && nvidia-smi -L > /dev/null 2>&1; then
+    nv_out="$(nvidia-smi --query-gpu=utilization.gpu,memory.used,memory.total --format=csv,noheader,nounits -i 0 2>/dev/null | tr -d ' ')"
+    IFS=',' read -r gpu_pct vram_used_mb vram_total_mb <<< "$nv_out"
+elif [ -e /sys/class/drm/card0/device/gpu_busy_percent ]; then
+    gpu_pct="$(cat /sys/class/drm/card0/device/gpu_busy_percent 2>/dev/null || echo -1)"
+    if [ -e /sys/class/drm/card0/device/mem_info_vram_used ]; then
+        vram_used_mb=$(($(cat /sys/class/drm/card0/device/mem_info_vram_used) / 1024 / 1024))
+        vram_total_mb=$(($(cat /sys/class/drm/card0/device/mem_info_vram_total) / 1024 / 1024))
+    fi
+fi
+
+# --- Disks (root + home) ---
+# Positional ($NF for mountpoint), not header-text matching — `df -h`'s
+# header is locale-translated (verified live: "System plików"/"%uż." under
+# this machine's pl_PL locale, not "Filesystem"/"Use%"), so matching on
+# English header text would silently find nothing.
+disks_json="$(df -h 2>/dev/null | awk '
+    NR == 1 { next }
+    $NF == "/" || $NF == "/home" {
+        gsub(/%/, "", $5)
+        printf "%s{\"mount\":\"%s\",\"size\":\"%s\",\"used\":\"%s\",\"avail\":\"%s\",\"pct\":\"%s\"}", (n++ ? "," : ""), $NF, $2, $3, $4, $5
+    }
+')"
+
+printf '{"cpuPct":%s,"memUsedKb":%s,"memTotalKb":%s,"gpuPct":%s,"vramUsedMb":%s,"vramTotalMb":%s,"disks":[%s]}\n' \
+    "$cpu_pct" "$mem_used_kb" "$mem_total_kb" "$gpu_pct" "$vram_used_mb" "$vram_total_mb" "$disks_json"