Add additional drop_likelihood_columns guards (#3333)

* Expose drop_likelihood_columns and apply it

Rename the private _drop_likelihood_columns to drop_likelihood_columns and use it across the codebase to consistently sanitize annotation DataFrames (drop any coord level named 'likelihood'). Updated imports/usages in modelzoo (ma_dlc, ma_dlc_dataframe), pose_estimation_pytorch (dlcloader), and utils (skeleton). Added a docstring note about centralizing project I/O and maintenance concerns. Added a unit test (tests/pose_estimation_pytorch/data/test_dlc_dataloader.py) to ensure DLCLoader.to_coco ignores likelihood columns.

* Warn and drop likelihood columns; normalize data var

Change logging from info to warning (with stacklevel=2) when likelihood columns are detected and drop them. Call drop_likelihood_columns immediately after reading the HDF file. Normalize variable naming by replacing Data with data throughout mergeandsplit and update subsequent references (index ranges, scorer extraction, and enumeration) to use the lower-case variable.