Implement new database for running simple tests

[cyberixae]

BSMR has a simple web filesystem (PetriFS) that is used for doing simple test runs. A slightly more complex database is needed for doing test runs that start from where the previous computation ended.

The input plugin for the new test database should read an entire directory of files. Feeding (filename, filedata) pairs to mappers. The output plugin for the directory needs to write the results into a directory by using the bucketId as the filename and the reducing result as the file contents. The factories that produce the plugins should take a dataset identifier as a parameter and select the correct directory as a result.

It should be possible to chain computations by giving the input plugin factory of a current run the dataset identifier that was given to the output plugin factory of a previous run. It should follow that the mapper input pairs of the second run are equal to the (bucketId, reducerOutput) of the first run.