fix: don't realign a window that is within a point of filling its zone - #1853

Fixes #1852.

Problem

With the Dock on the right, Maximize leaves the window 1pt narrower than the screen and shifted 1pt right (x=1 instead of x=0). From the log in the issue, on a 1679pt-wide visible frame:

AX sizing proposed: (1679.0, 1079.0), result: (1678.0, 1079.0)
AX position proposed: (1.0, 38.0), result: (1.0, 38.0)
maximize, calculatedRect: (0.0, 38.0, 1679.0, 1079.0), resultRect: (1.0, 38.0, 1678.0, 1079.0)

Two things combine, as diagnosed by @nyg in the issue:

1. macOS shortens an AX size change that grows a window's right edge onto the Dock edge, so a requested width of 1679 comes back as 1678. A position change to the same edge is not clamped, which is why Left Half → Last Two Thirds (resized at x=0, then moved) is exact while Right Half → Last Two Thirds is not.
2. ClampedWindowAligner reads that 1pt shortfall as a clamped window. Maximize shares both left and right screen edges, so the window gets centered: round((1679 - 1678) / 2) = 1.

The net effect is that the 1pt gap moves off the Dock edge, where it was invisible, onto the opposite edge, where it is not.

Fix

Skip realignment on an axis when the shortfall there is within a point. I took the first of the two options suggested in the issue — the alternative, applying the size to the left of the target and then moving the window, would change AccessibilityElement.setFrame behavior for every action to address a 1pt cosmetic issue.

The check is per axis, so a window that is a point narrow but genuinely short still gets aligned vertically. Real clamps — a window holding a fixed aspect ratio, for example — are unaffected, since those miss the zone by far more than a point.

Tests

Five cases added to ClampedWindowAlignerTests, using the measurements from the issue:

• testMaximizeOnePointNarrowStaysPut — the reported case (zone 1679 wide, window 1678, all four edges shared)
• testRightHalfOnePointNarrowStaysPut — the right-half variant from the issue's measurement table
• testOnePointShortHeightStaysPut — same tolerance on the vertical axis
• testOnePointNarrowLeavesOtherAxisAlignmentIntact — a point narrow and genuinely short: width is left alone, height still centers
• testShortfallJustOverToleranceStillAligns — pins the boundary so the tolerance can't be widened later without a test failing

The first four fail on main and pass with this change. The boundary test passes both ways by design. Full suite: 338 tests, 0 failures.