perf(ui/state): cache lowercase sort key for grouped-process ordering

[obchain]

Summary

compute_grouped_rows sorted process groups with:

group_stats.sort_by_key(|a| a.0.to_lowercase());

slice::sort_by_key does not cache the extracted key — it calls the closure on every comparison, so a merge sort invokes it O(N log N) times. Each call allocates a fresh lowercase String, so sorting N process groups performs ~N·log₂(N) heap allocations where N would do.

compute_grouped_rows runs on every regroup in the Overview grouped view (toggling a, plus refreshes while grouping is active), so this recurs on the update path, not just once.

Switch to slice::sort_by_cached_key, which calls the key function exactly once per element, caches the results, and sorts those. Same key, same ordering — strictly fewer allocations. For 64 process groups that's 64 lowercase Strings per regroup instead of ~384.

Closes #352.

Behavior

No change — identical sort key and comparison. Verified by:

cargo test --lib                          # 379 passed
cargo fmt --check                          # clean
cargo clippy --all-targets -- -D warnings  # clean

Test plan

• Full cargo test --lib
• cargo fmt --check
• cargo clippy --all-targets -- -D warnings