Add standardised and down-sampled result file

Currently, the output of bilby_pipe can be confusing to users. If n-parallel=1, they get a single result file. But, if n-parallel>1, they get n-parallel result files + a merge file (which is the "final" data product).

In addition, the result files can be very large (several hundred Mb) if combining multiple runs.

To resolve both issues we could

• Add a directory {outdir}/output/{label}_final_result.json where {label} would be the full label (including gpstime etc)
• Add an option max-samples=30000 which would thin the original sample set (preserving the order) until the number of samples is less than max-samples

Thoughts?