Add a SafeMaskMaker at DL3 level

[AtreyeeS]

This PR implements a SafeMaskMaker to run on DL3 IRFs as discussed in #3878 .
The cuts are applied directly on the event list.
Since HAWC and SWGO are directly using DL4 IRFs, I guess there is no need to adapt this PR for non pointing instruments.

In this case, the SafeMakerDL3 must be run before the MapDatasetMaker, as opposed to the SafeMaskMaker

The energy dispersion is not applied as yet.

[registerrier]

Thanks @AtreyeeS . This is interesting, but I think we have to understand how this fits in the workflow.

A few general comments/questions:

• technically this is more an event filtering scheme than a mask creation algorithm. One could use the existing ObservationFilter.
• one can probably split event filtering from mask creation, but the latter is still necessary for further treatment and analysis.
• the advantage of creating the masks on the input geometry and apply them to filter events is that you don't have to worry about fractional exposures in the bins. If not, mask_safe has to contain weights between 0 and 1 rather than booleans. This might be useful in a number of situation but requires heavy modifications in stacking and other functionalities. Currently edisp is the place where we do this (and once stacked, the mask_safe is partly redundant with the EdispKernelMap).
• The idea of having event filtering as part of the list of Makers is interesting because it could allow reading observations from the DataStore within the loop over observations (or in that case obs_id).

One option could be first to adapt the SafeMaskMaker to use the observation IRF rather than the Dataset IRF to build the mask. This would alleviate some of the difficulties we have currently.

Opinions @adonath @QRemy ?

[codecov]

Codecov Report

Attention: 61 lines in your changes are missing coverage. Please review.

Comparison is base (921f1a9) 75.71% compared to head (8b75dc5) 75.67%.
Report is 96 commits behind head on main.

Files	Patch %	Lines
gammapy/makers/safe.py	4.68%	61 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4599      +/-   ##
==========================================
- Coverage   75.71%   75.67%   -0.05%     
==========================================
  Files         228      229       +1     
  Lines       33778    33865      +87     
==========================================
+ Hits        25576    25627      +51     
- Misses       8202     8238      +36     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[AtreyeeS]

This should be ready for review now

[QRemy]

Thanks @AtreyeeS, I left few minor comments. It may be further simplified to have less nested if tests, but if it is not possible it may be nicer to move the DL3/DL4 cases in separate functions. I will have a closer look into that.

Edit: I removed a bit of code duplication, @AtreyeeS, @registerrier could you look back to it ?

[registerrier]

Thanks @AtreyeeS . This looks good. I have only minor comments.

[registerrier]

Suggested change

	raise IOError("Invalid option fro irfs. Choose 'DL3' or 'DL4'.")
	raise ValueError("Invalid option for irfs: expected 'DL3' or 'DL4', got {irfs} instead.")

[registerrier]

Suggested change

	Whether to use reprojected ("DL4") or raw ("DL3") irfs
	Default is "DL4"
	Whether to use reprojected ("DL4") or raw ("DL3") irfs.
	Default is "DL4".

[AtreyeeS]

Thanks @registerrier ! Commited your suggestions. Should be good to merge now

[registerrier]

Thanks @AtreyeeS . No further comment.

[registerrier]

CI fail is unrelated. Merging now.