Honor screenEdgeGapTopNotch when it is set to 0

[efim0v]

Fixes #1658.

Problem

screenEdgeGapTopNotch was a FloatDefault, which reads UserDefaults.float(forKey:). That returns 0 when the key is missing, so a missing key and an explicit 0 looked the same. adjustedVisibleFrame used value != 0 to mean "set", so screenEdgeGapTopNotch = 0 fell back to screenEdgeGapTop on screens with a notch. With screenEdgeGapTop = 30, the only way to get no top gap on the built-in display was a tiny non-zero value such as -0.00001.

Fix

• Add OptionalDoubleDefault, modeled on DoubleDefault. Its value is nil when the key is missing, and setting it back to nil removes the key.
• Use it for screenEdgeGapTopNotch. On a screen with a notch, any set value, including 0, replaces screenEdgeGapTop. When it is not set, screenEdgeGapTop is used as before. Non-zero values and screens without a notch behave as before.
• Config export writes a set value as double and an unset value as an empty entry ({}). Importing the empty entry clears the setting, so "unset" and 0 stay distinct through export and import.
• Configs exported by earlier versions contain "screenEdgeGapTopNotch": {"float": 0} even when it was never set, so a legacy float 0 is still imported as "not set", following how DoubleDefault handles legacy float values (Allow zero to disable minimum window size limits #1802). Other legacy values are imported as they are.
• TerminalCommands.md now describes the 0 case and how to go back to the fallback with defaults delete.

One behavior change to be aware of: importing one of those older configs used to store 0 for this key. Someone who imported such a config before this change, has screenEdgeGapTop set and uses a notched display will get no top gap on that display after updating. defaults delete com.knollsoft.Rectangle screenEdgeGapTopNotch restores the fallback.

Tests

Added ScreenEdgeGapTests. It calls adjustedVisibleFrame() on NSScreen test doubles with and without a top safe-area inset, with screenEdgeGapTop = 30:

• notch gap unset: 30 on both screens
• notch gap 0: 0 on the notched screen, 30 on the other. On main this fails with XCTAssertEqual failed: ("30.0") is not equal to ("0.0").
• notch gap 10: 10 on the notched screen, 30 on the other
• unset and 0 both survive reloading the default and a config export/import
• a legacy {"float": 0} entry imports as not set (30 on the notched screen), and a legacy non-zero value is used

I ran the full RectangleTests bundle against a Debug build (Xcode 27.0, macOS 27.0): 572 tests, and the new tests pass. Some existing tests fail in my setup on both main and this branch, so they are unrelated to this change: ShortcutRecordingObserverTests.testRecordingAMouseButtonEndsRecording and two tests in WindowActionShortcutViewTests fail on every run, and three tests in SnappingManagerSessionTests fail in some runs (they read the real front window's full-screen state).

[rxhanson]

Thanks for working on this problem! I think this could also be solved by using a DoubleDefault with a default value of, say, -1 since DoubleDefault applies the default value only if the object is nil? The benefit of doing it that way is we don't have to introduce yet another default class, although it does make the usage a little awkward. If it does make more sense to keep the OptionalDoubleDefault, CGFloat and Double are the same thing, so you would simply read the double value of it.

[efim0v]

Good call, that's simpler. Switched to DoubleDefault with a default of -1 and dropped OptionalDoubleDefault: a negative value means it isn't set, so screenEdgeGapTop applies, and 0 now works. Older exported configs that stored 0 for this key still import as not set.