It may very time-consuming to feed a large number of files with relations 
(e.g., coming from distributed sieving) to `procrels` since
* `procrels` accepts only one file with new relations per run;
* it takes long time for `procrels` to read previously processed relations.
So it seems intractable even to feed 2000 files to `procrels` assuming 
that each run takes 5 minutes.

`relsort` is a simple pre-processor of files with relations that re-packs 
them into a smaller number of files.

As an input it takes a prefix of the name of input files and a prefix of 
the name of output files. The prefixes may include path to the files. 
Maximum size of the output files is defined by constant `limit` in the 
sources (currently, it is 2GB).

It outputs to stdout shell script like:
cat infile1 infile6 ... >outfile1
cat infile5 infile8 ... >outfile2
...

Running this script will re-pack large number of input files infile* into a smaller
number of output files outfile*.
