BalancedPartitioning: minor updates (#77568)

When LargestTraceSize is a power of two, createBPFunctionNodes does not
allocate a group ID for Trace[LargestTraceSize-1] (as N is off by 1).
Fix
this and change floor+log2 to Log2_64.

BalancedPartitioning::bisect can use unstable sort because `Nodes`
contains distinct `InputOrderIndex`s.

BalancedPartitioning::runIterations: use one DenseMap and simplify the
node renumbering code.