Uncertainties in IRF and IRF metric output

[kosack]

Please describe the use case that requires this feature.

The outputs of ctapipe-compute-irfs should include uncertainties when possible, particularly for performance metrics output by the --benchmark-output option. This is critical for comparisons, especially where we have low event statistics.

Describe the solution you'd like

Include an extra column in the output tables for the uncertainty. This is a data model change (an addition), but shouldn't interfere with existing GADF tools.

For example, in a EFFECTIVE AREA bintable we have columns:

  name    dtype   shape  unit
-------- ------- ------- ----
ENERG_LO float64   (42,)  TeV
ENERG_HI float64   (42,)  TeV
THETA_LO float64    (6,)  deg
THETA_HI float64    (6,)  deg
 EFFAREA float32 (6, 42)   m2

And could just add a column EFFAREA_UNCERT with the statistical uncertainties. For that one, they are easy to compute using simple error propagation. For sensitivity, etc, it's a bit more complex, but still doable.

Note this may require a complimentary issue for pyirf.