Fix: exclude kernel threads from process enumeration to prevent RSS double-counting

[RobertoMaurizzi]

Summary

Recent versions of sysinfo (0.30+) enumerates /proc/[pid]/task/[tid] entries as separate processes. On Linux each thread's /proc/[tid]/statm returns the full parent-process RSS (all threads share mm_struct), so a 30-thread browser inflates the per-process RSS sum 30×. This makes the treemap grouping denominator (total_value) far larger than actual physical memory, causing every entry to fall below group_threshold (default 1%) and collapse into a single "Other" tile.

The fix uses ProcessRefreshKind::without_tasks() on both the initial and refresh-specifics calls, which tells sysinfo to skip scanning /proc/[pid]/task/ entirely. This is both cleaner and more efficient than post-filtering — no thread entries are created, no wasted I/O reading their stat files.

Before the change

After the change

Validation

• cargo fmt -- --check
• cargo clippy --all-targets --all-features -- -D warnings
• cargo test

Notes

I'm not sure if there are cases where somebody might want to run actually counting threads: my main doubt that makes me think this is the right fix is that since each thread returns the whole RSS then there's no point trying to display them in a treemap showing memory usage distribution.
If you see a use for that, I might instead collect the threads but add a toggle about how they are managed.

I'm unable to test this on Windows or MacOS.