Fix fragged rows signed-ness regression

[tobil4sk]

Fixes a behavioural regression in 952b6ac, which caused the GC to continue allocating more blocks due to sWorkingMemorySize becoming corrupted.

The problem is that if spaceStart is greater than spaceEnd, spaceEnd - spaceStart would give a very large size_t instead of a negative int, and therefore frag > 0 was true and mFraggedRows got updated to a random value. This had a domino effect on various variables before corrupting sWorkingMemorySize.

hxcpp/src/hx/gc/Immix.cpp

Lines 6293 to 6296 in f05b21c

	size_t frag{ spaceEnd - spaceStart };
	if (frag>0)
	*mFraggedRows += int{ static_cast<int>(frag >> IMMIX_LINE_BITS) };
	}

hxcpp/src/hx/gc/Immix.cpp

Line 877 in f05b21c

outStats.fraggedRows += mFraggedRows;

hxcpp/src/hx/gc/Immix.cpp

Line 5043 in f05b21c

mRowsInUse = stats.rowsInUse + stats.fraggedRows + freeFraggedRows;

hxcpp/src/hx/gc/Immix.cpp

Line 5225 in f05b21c

size_t mem = mRowsInUse<<IMMIX_LINE_BITS;

hxcpp/src/hx/gc/Immix.cpp

Line 5235 in f05b21c

sWorkingMemorySize = std::max( mem + targetFree, (size_t)hx::sgMinimumWorkingMemory);

hxcpp/src/hx/gc/Immix.cpp

Lines 3607 to 3611 in f05b21c

	if (!result && allowMoreBlocks() && (!sgInternalEnable || GetWorkingMemory()<sWorkingMemorySize))
	{
	if (AllocMoreBlocks(forceCompact,false))
	result = GetNextFree(inRequiredBytes);
	}

I haven't looked into whether it even makes sense for spaceStart to be bigger than spaceEnd, but for now this brings back the old logic by skipping the update in that case.