Patch cross-section scaling when filtering weight streams

When filtering weight-streams at run time (--match-weight, --unmatch-weights, --nominal-weight), the wrong cross-sections might be picked up in finalize. This change set fixes this and also streamlines the behaviour of --nominal-weight which supplies a custom nominal weight name.

The logic is as follows:

• Rivet runs its usual heuristic to identify the nominal weight stream
• If it finds more than one, it will use the first one and print a warning that there were potentially other candidates, thereby giving the option to abort and re-run using a suitable --nominal-weight value.
• If it can't find one (that's a bug in the generator HepMC interface ...), it checks if the user has supplied a custom weight name that ought to be treated as the nominal and uses that instead
  • If the user didn't supply one, Rivet falls back to the assumption that the first weight is the nominal (technically a HepMC convention) but will write out all weights streams in "variation weight" style (i.e. with [weight_name] suffix in the path). A warning is printed.
• If it can find the nominal weight stream and the user supplied a custom one that isn't the actual nominal weight stream, Rivet will internally continue to use the actual nominal (e.g. for the cross-section scaling) but write out the user-supplied weight stream in "nominal weight" style (and the actual nominal in "variation weight" style)

If the user runs with the --skip-weights flag:

• All weights, except for the actual default weight stream are unmatched. Only one weight stream is written out to file (in "nominal weight" style).
• If the user additionally supplied --match-weight and --unmatch-weight flags, it's probably safe to assume they don't know what they're doing. Preference is given to --skip-weights in this case.
• If the user also supplies the --nominal-weight flag and asks that a variation weight be treated as the nominal, two weight streams are written out to file: the actual default weight stream (needed for scaling) and the custom nominal weight stream. The former will be written out in "variation weight" style, the latter in "nominal weight" style.