Merge `NoLabelDuplicates` And `LinkedParameters` Constraints Into `DegeneracyConstraint`

[Scienfitz]

Motivation

DiscreteNoLabelDuplicatesConstraint and DiscreteLinkedParametersConstraint are two extremes of the same degeneracy concept (value repetition across a parameter group). They can be unified into a single DiscreteDegeneracyConstraint parametrized by n_max_occurrences (the maximum number of times any single value may appear in a row), combined with the inherited exclude flag.

• n_max_occurrences=1 → keep only all-distinct rows (old DiscreteNoLabelDuplicatesConstraint)
• n_max_occurrences=M-1, exclude=True → keep only all-identical rows (old DiscreteLinkedParametersConstraint, with M the number of parameters)

Todo

• Introduce DiscreteDegeneracyConstraint with n_max_occurrences: int = 1 (default = all-distinct), reusing the inherited exclude flag.
• Implement pandas and Polars matching-row logic (row kept iff no value occurs more than n_max_occurrences times).
• Deprecation (per AGENTS.md §10): replace the two old classes with same-named factory functions that warn and return the correspondingly configured DiscreteDegeneracyConstraint; add cattrs (de)serialization redirects for the old class names; cover in tests/test_deprecations.py.
• Replace the two entries in the pruning order tuple with DiscreteDegeneracyConstraint.
• Update fixtures, hypothesis strategies, docs, and examples.
• [ ]

Code Refs

• DiscreteNoLabelDuplicatesConstraint: baybe/constraints/discrete.py:220-260 (pandas + Polars)
• DiscreteLinkedParametersConstraint: baybe/constraints/discrete.py:263-297 (pandas + Polars)
• Deprecation pattern to mirror (`DiscreteExclu

[Scienfitz]

Update:

the initial plan was. wrong. The main issue is that DiscreteNoLabelDuplicatesConstraint and DiscreteLinkedParametersConstraint are not simple complements of each other.

However it is still possible to combine both in a single new constraint that operates on how many unique values are allowed. Plan:

• DiscreteDegeneracyConstraint (name still open for discussion)
• n_max_duplicates which defaults to 1

The two old constraints can the imply be recovered by using

• DiscreteDegeneracyConstraint(n_max_duplicates=1) -> same as DiscreteNoLabelDuplicatesConstraint
• DiscreteDegeneracyConstraint(n_max_duplicates=M-1, exclude=True) where M is the number of affected parameters -> same as DiscreteLinkedParametersConstraint

This choice is also motivated by recognizing that DiscreteNoLabelDuplicatesConstraint is much more important and more frequently used, DiscreteLinkedParametersConstraint is exotic