Add BCRSMatrix::setIndicesNoSort() and speedup MatrixIndexSet::exportIdx()

So far MatrixIndexSet::exportIdx() used BCRSMatrix::addindex() on individual column indices. This was slow because each insertion does a binary search although the inserted indices are already sorted. Bulk-inserting whole rows with setIndices() improves on this significantly but still does a non-necessary sort. The latter is avoided by the new BCRSMatrix::setIndicesNoSort() method. This may speedup exportIdx() significantly (e.g. by factor 2 for Poisson with PQ2).

In general exportIdx() is already cheap compared to building the pattern in typical matrix assembly. But since exportIdx() is serial, the improvement becomes significant in multi-threaded assembly.

According to my measurements exportIdx() is dominated by the allocation after this patch, such that there's not much potential for further improvements.