Inference center correction

[VincentB03]

Changes:

• Removed the offset from scene.py::build_batched_model on x and y (+0.5) which would result in a discrepancy between the inferred position of the galaxy's center and the observed position

• Added a unit test (test_centers_match) to verify that the positions of the centers of the observed and inferred galaxies are the same in test_btached_inference::TestBuildBatchedModel. This test reuses the configuration used in lvl0 and runs a short inference to verify that the centers of the observations and the centers recorded during inference are the same

• Test workflows have been restricted to Python 3.11 (3.12 and 3.13 have been removed) due to compatibility issues with the Arviz library : """ FAILED tests/test_validation/test_extraction.py::TestExtractConvergenceDiagnostics::test_rhat_near_one - TypeError: from_dict() got an unexpected keyword argument 'posterior'
  """ This new library version can only be installed on Python 3.12 or later
  Migration guide : https://python.arviz.org/en/latest/user_guide/migration_guide.html

[sfarrens]

Have you tested that everything works if you change the image sizes and, importantly, change from an even number to and odd number of pixels?

[VincentB03]

Have you tested that everything works if you change the image sizes and, importantly, change from an even number to and odd number of pixels?

I tried 49x49, 48x48, 25x25 and 64x64, by selecting the x_min/x_max and y_min/y_max values that allow for accurate inference each time, the results are no longer biased.