Improving CUnitQueue performance: don't adjust mcount

[maxsharabayko]

Extracted undisputed changes from PR #2395.

Do not adjust taken units in CUnitQueue::increase()

The re-estimation of the number of units taken in the CUnitQueue was previously done as a workaround for synchronization issues around simultaneous access to CUnitQueue::m_iCount.
Those issues should be fixed by now, e.g. #1333.
No point in this CPU-heavy loop anymore.

while (p != NULL)
{
	CUnit* u = p->m_pUnit;
	for (CUnit* end = u + p->m_iSize; u != end; ++u)
		if (u->m_iFlag != CUnit::FREE)
			++real_count;

	if (p == m_pLastQueue)
		p = NULL;
	else
		p = p->m_pNext;
}

CRcvQueue now tracks IP version instead of CUnitQueue

Previously CUnitQueue was tracking IP version. The value was submitted in CRcvQueue::init(..), and used by CRcvQueue::worker(..) thread. There is no use for the value in the CUnitQueue itself.
Therefore, moved to CRcvQueue.

Minor clean-up fix rethrowing an exception.

An exception was being caught in a named variable, while it was not referenced.