perf: skip the shortcut walk for modifier sets no shortcut uses

[arzafran]

What this does

Typing capital letters, shifted symbols and Option characters gets cheaper per key. Those key presses used to check all 57 shortcut actions one by one, even though no shortcut uses Shift alone or Option alone. Now the app keeps the set of modifier combinations that any shortcut starts with, and skips the whole check when a key press uses a combination outside that set.

Plain typing is unchanged: it already skipped the check.

How it stays safe

• Every shortcut matcher compares modifiers first, so a key whose modifiers aren't in the set can't match anything.
• The set includes the hardcoded Ctrl+Tab and Ctrl+Shift+Tab, and the Ghostty split-navigation shortcuts are checked live.
• A pending two-key chord always runs the full check.
• The set is rebuilt whenever shortcuts change: in Settings, by reset, or by a settings.json reload. Until it is first built, nothing is skipped.

Measured

Debug build, 20 batches of 1000 synthetic key presses, median per key, 3 runs each (numbers are from the whole shortcut handler).

Key	Before	After
Shift+A	165-166 µs	66-69 µs
Shift+:	162-220 µs	67-68 µs
Option+A	169-220 µs	67-69 µs
Cmd+Shift+Y, Ctrl+K (shortcut modifiers)	unchanged	unchanged

Debug numbers overstate absolute cost; the ratio is the point. The benchmark harness is not included.

Test plan

• AppDelegateShortcutRoutingTests pass in CI.
• Capital letters, symbols and Option characters type normally in a terminal and the browser address bar.
• A shortcut changed to an unusual modifier set in Settings works right away without restarting.