Adding a metadata in the datasets precising the origin of the background data

[bkhelifi]

Is your feature request related to a problem? Please describe.
As illustrated in the issue #5245, some argument of estimators might be not compatible with the type of background, how it is built.
In parallel, as the dataset background cubes are a set of data processed by Gammapy, its origin should be described.

Describe the solution you'd like
The global scheme could be :

1. a dataset metadata precising the background type/origin (e.g. fits_keyword = BkgOrigin' or BkgType' and values=enum(None, IRF, ScaledIRF, Ring, Reflected, OFF)
2. provenance data that gives the maker parameters used to compute it (allowing on the long-term the reproducibility).

Additional context

• Depending of the exact needs of the estimators, the used solution might need to be adjusted.
• Once the scheme is decided, the metadata could be added into the 'Context Metadata' of the `Dataset' class. And then, the BackgroundMaker should fill correctly this metadata. Finally, the information should be correctly propagated into the dataflow of the data reduction (see the project https://github.com/orgs/gammapy/projects/13)