Improve Offset dependence in IRFBenchmark

Right now the offset dependence in the IRF benchmark has a few issues and missing features:

1. If the offset bins don't match between InputDatasets, the plots are not generated. This is maybe not a solvable problem, and we should try to use the same offset bins when producing inputs, but having the option to e.g. "use the lowest offset bin" and ignore others would be nice for already-produced datasets that do not match exactly. This is particuarly useful since often we want to compare to the published curves, which means we have to follow their bining. Better error messages are also needed, as it's not clear to the user why the plots don't appear when bins are mismatched.
2. We get a PosterPlot for each offset (good). However, it would also be useful to have the inverse: for each MetricStore, generate the PosterPlot with the lowest offset as reference, and compared to all other offsets. That allows a second way to see offset dependence.

E.g this plot fails if the any of the MetricStores being compared have different offset bin edges:

When that happens, no output is produced, and the error is only:

Background IRF does not have offsets this far: 1.0 to 2.0 deg

It's not clear which MetricsStore is at fault, and what the binning should be.