Add top-processes drawer to the CPU widget

Clicking the CPU pill opens a popup listing the highest-CPU processes,
refreshed live while open. services/topproc.sh computes top-style per-process
CPU% and mem% from two /proc snapshots and prints one ranked frame per run;
SysStats streams it only while procPollEnabled (drawer open), so the bar pays
nothing at rest.

CLAUDE.md documents the architecture and the runtime constraints discovered
here (helper scripts must be git-tracked for `nix run`; single-shot + exit to
flush stdout; bash+coreutils only; StdioCollector/Timer don't fire in this
Quickshell build; fixed-size popup to avoid stale-buffer on resize under Sway).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

services/topproc.sh

@@ -0,0 +1,79 @@
+#!/usr/bin/env bash
+# Print one frame of the top CPU-consuming processes, then exit.
+#
+# Single-shot by design: SysStats re-runs it every procInterval while the CPU
+# drawer is open (Process.running re-armed by a Timer, like dfProc). Exiting
+# per frame guarantees stdout is flushed — a long-running loop kept its frame
+# delimiter stuck in bash's block buffer and the QML side never saw a full
+# frame. discover.sh is the reference for the bash-script + parse pattern.
+#
+# IMPORTANT: the flake wrapper only puts bash + coreutils on PATH, so this uses
+# ONLY bash builtins and coreutils (nproc, sort, head, sleep, cat) — no awk, no
+# getconf. Keep it that way or add the tool to runtimeInputs in flake.nix.
+#
+# Output: one "P <cpu%> <mem%> <name>" line per process, highest CPU first.
+# cpu% is top-style: 100 == one fully-used core.
+
+intv="${1:-2}"
+n="${2:-6}"
+ncpu="$(nproc 2>/dev/null || echo 1)"
+pg_kb=4         # page size is 4096B on all targeted arches (x86_64/aarch64)
+
+# MemTotal in kB, read without awk.
+memkb=0
+while read -r key val _; do
+    if [ "$key" = "MemTotal:" ]; then memkb="$val"; break; fi
+done < /proc/meminfo
+
+# Aggregate CPU jiffies: sum of the fields on the leading "cpu" line.
+cpu_total() {
+    local line f s=0
+    read -r line < /proc/stat
+    # shellcheck disable=SC2086
+    set -- $line            # collapses the double space; $1 == "cpu"
+    shift
+    for f in "$@"; do s=$(( s + f )); done
+    echo "$s"
+}
+
+# Read pid -> (utime+stime) jiffies and rss pages into the given assoc arrays.
+# /proc/<pid>/stat is "PID (comm) STATE ...": strip through ") " so comm spaces
+# and parens can't shift field positions, then utime=14, stime=15, rss=24.
+snapshot() {
+    local -n _jif="$1" _rss="$2"
+    local f l pid rest
+    for f in /proc/[0-9]*/stat; do
+        # 2>/dev/null first so a process vanishing mid-scan is silent.
+        IFS= read -r l 2>/dev/null < "$f" || continue
+        pid="${l%% *}"
+        rest="${l#*) }"
+        # shellcheck disable=SC2086
+        set -- $rest
+        _jif["$pid"]=$(( ${12:-0} + ${13:-0} ))
+        _rss["$pid"]="${22:-0}"
+    done
+}
+
+declare -A j1 j2 rss
+t1="$(cpu_total)"
+snapshot j1 rss
+sleep "$intv"
+t2="$(cpu_total)"
+snapshot j2 rss
+dt=$(( t2 - t1 ))
+[ "$dt" -gt 0 ] || exit 0
+
+# Emit "<cpu> <pid> <rss>" for each process seen in both snapshots, sort by cpu
+# desc, take the top n, then resolve name + mem% for just those.
+for pid in "${!j2[@]}"; do
+    [ -n "${j1[$pid]:-}" ] || continue
+    d=$(( ${j2[$pid]} - ${j1[$pid]} ))
+    cpu=$(( 100 * ncpu * d / dt ))
+    printf '%d %s %s\n' "$cpu" "$pid" "${rss[$pid]:-0}"
+done | sort -rn | head -n "$n" | while read -r cpu pid r; do
+    name="$(cat "/proc/$pid/comm" 2>/dev/null)"
+    [ -n "$name" ] || name="$pid"
+    mem=0
+    [ "$memkb" -gt 0 ] && mem=$(( r * pg_kb * 100 / memkb ))
+    printf 'P %s %s %s\n' "$cpu" "$mem" "$name"
+done