Benchmark context collections of different sizes

The benchmark should use collections of different size in order to measure the effect of "noise" on context retrieval. For this, the benchmark tool should support assigning a document to several collections.