Fix findOutline4 emitting interior mask pixels as outline pixels

[rootkiller6788]

Problem

MaskAnalyzer::findOutline4 (impl/ocean/cv/segmentation/MaskAnalyzer.cpp) documents its result as: "the pixel itself is not a mask pixel but has at least one neighbor mask pixel in the four-neighborhood" (see MaskAnalyzer.h).

However, the center-rows pass only checked the four-neighborhood of each pixel:

if (maskRow[x - 1u] != nonMaskValue || maskRow[x + 1u] != nonMaskValue || maskRowTop[x] != nonMaskValue || maskRowBottom[x] != nonMaskValue)
{
    outlinePixels4.emplace_back(x, y);
}

It never verified that the pixel itself is a non-mask pixel. As a result, every interior mask pixel that has at least one mask neighbor is incorrectly reported as an outline pixel, so outlinePixels4 contains the full mask interior in addition to the true outline.

Fix

Gate the check on the pixel being a non-mask pixel, matching the top/bottom row and left/right column passes already present in the same function:

if (maskRow[x] == nonMaskValue && (maskRow[x - 1u] != nonMaskValue || maskRow[x + 1u] != nonMaskValue || maskRowTop[x] != nonMaskValue || maskRowBottom[x] != nonMaskValue))
{
    outlinePixels4.emplace_back(x, y);
}

Verification

• Confirmed the header documentation (MaskAnalyzer.h:459) defines outline-4 pixels as non-mask pixels with a mask neighbor.
• Confirmed the top-center (line 373/383) and bottom-center (line 513/523) passes already use the maskRow[x] != nonMaskValue / else structure, i.e. they correctly skip mask pixels.
• Traced a concrete counterexample: a fully-mask 3x3 frame produces an outline entry for the interior pixel (1,1), which is a mask pixel and must not be an outline pixel.