Address review feedback (@rschu1ze): add an .xml performance test that
demonstrates the speedup of the bulk single-copy path for trivially
serializable vectors.
The test exercises writeVectorBinary / readVectorBinary through the
analysisOfVariance aggregate state, whose state holds three large contiguous
vectors (std::vector<Float64> and std::vector<UInt64>). Writing the
AggregateFunction column to MergeTree serializes the state; reading it back
deserializes it.
