Fix WithinSession/WithinSubjectSplitter overwriting explicit n_splits in cv_kwargs

[Copilot]

WithinSessionSplitter and WithinSubjectSplitter unconditionally injected n_folds/shuffle/random_state into the inner CV kwargs, clobbering values the caller explicitly passed via cv_kwargs. This made it impossible to request, e.g., a single stratified holdout split without an inner-CV wrapper that hides n_splits.

from sklearn.model_selection import StratifiedShuffleSplit
from moabb.evaluations.splitters import WithinSessionSplitter

# Previously gave 5 folds (n_splits clobbered by n_folds); now yields one 75/25 split per session
sp = WithinSessionSplitter(cv_class=StratifiedShuffleSplit, n_splits=1,
                           test_size=0.25, shuffle=True, random_state=42)

Changes

• moabb/evaluations/splitters.py — In both splitters, only inject the n_folds/shuffle/random_state defaults when the key is absent from cv_kwargs (if p in params and p not in cv_kwargs). Caller-provided values now take precedence; defaults are unchanged when not supplied.
• Docstrings — Note that explicit n_splits/shuffle/random_state in cv_kwargs override the corresponding constructor arguments.
• moabb/tests/test_splits.py — Parametrized regression test asserting an explicit n_splits=1 survives and produces one split per group for both splitters.
• docs/source/whats_new.rst — Changelog entry.