fix(command-mode): detect destructive commands in chained segments

[postoso]

What

CommandModeService.isDestructiveCommand applied its prefix checks to the whole command string, so a destructive command hidden after a shell chaining separator bypassed the confirm gate and auto-ran. Examples that returned false (auto-run): echo ok && killall Finder, true; shred secret.txt, echo a || killall Finder, echo ok & killall Finder, and a newline-chained sudo reboot.

How

Split the command on chaining separators (&&, ||, ;, &, newline) into segments, trim each, and apply the existing detection per segment (factored into isDestructiveSegment). && is normalized before the single & so it is not read as two background separators. Pipe (|) is intentionally not a split point: piped destructive commands stay matched by the existing | rm / | sudo / | dd substring patterns. Verified that 2>&1 and sleep 10 & echo done do not false-positive.

Tests

New CommandModeDestructiveCommandTests (15 methods): each separator hiding a destructive command, safe chains, a backgrounded-safe command, and preserved single-command plus piped detection. xcodebuild test is green and swiftlint --strict is clean.

Note

Focused follow-up to #428 (which hardens the same function for pipe-to-shell and output redirects). The two address independent gaps; this is structured so it rebases cleanly after #428 merges (the new per-segment isDestructiveSegment helper is where #428's added patterns belong).