Add background subpackage: CMD histogram resources and injection pipe…

[MatthieuPE]

Summary

Introduces the streamobs.background subpackage, providing two pipelines for
generating star and galaxy backgrounds alongside stream injections.

Survey / injector changes

• Add gal_misclassification field and get_gal_misclassification() to Survey
  (detection × misclassification probability for galaxies, without the bright-end
  1-padding applied to stars).
• Add SurveyFactory.set_gal_misclassification() and auto-loading from the
  gal_misclassification config key.
• Add source_type='stars'|'galaxies' parameter to StreamInjector.inject() and
  detect_flag().
• Add load_catalog() utility to utils.py (accepts a DataFrame or a parquet/CSV path).

New streamobs/background/ subpackage (scaffold — all methods stubbed)

• BackgroundCatalogInjector: thin wrapper around StreamInjector with
  inject_stars() / inject_galaxies() and a unified _prepare_survey()
  (uniform maglim, no-dust options).
• BackgroundResourceBuilder: builds 2-D color–magnitude histogram grids by
  injecting true catalogs into uniform surveys at a grid of (maglim_ref, Δmaglim)
  pairs.
• BackgroundStorage: saves/loads CMD grids as long-format zstd-compressed
  parquet files under data/background/ (gitignored). Exposes get_path(),
  save_data(), load_data(), exists().
• LightBackgroundGenerator: per-pixel fast generation — nearest-neighbour CMD
  lookup, Poisson count scaling, position sampling.
• Background: top-level wrapper dispatching to the full injection or light
  method; defaults to bundled package resources when no storage is provided.

Tests / docs

• tests/test_background.py: 31 test stubs marked @pytest.mark.background,
  using tmp_path for all disk I/O.
• docs/source/background.md (user guide) and
  docs/source/build_background_resources.md (developer guide).
• background marker registered in pytest.ini; data/background/ added to
  .gitignore.

[MatthieuPE]

So far coded:

• loading missclassification efficiency
• using it in the injection for galaxies if specified
• build missclassification + detection if only missclassification loaded
• StreamInjector wrapper to inject either stars or galaxies background

Now we should be able to get background realization from any truth catalogs.

Next:

• verify that nothing broken by visual inspection of the different efficiencies functions
• add tests for missclassification behavior
• Continue to fill the other parts of the code

WARNING
Missclassication and optionnaly missclassication+detection must be added to the data base of each survey. Now tests ran with missclassication+detection of lsst added by hand in the files