[ENH] Add OnlineBatchMixture and OnlineBootstrapRemember

[ravjot07]

[ENH] Implement online probabilistic regression meta-algorithms

This PR implements two online probabilistic regression meta-algorithms as specified in issue #464:

1. OnlineBatchMixture: Fits separate copies of regressor on batches and returns Mixture distributions with weights proportional to batch sizes
2. OnlineBootstrapRemember: Maintains a bootstrap sample from fit data and pools it with new batches, returning Mixture distributions

Closes #464

Changes

New Files

• skpro/regression/online/_batch_mixture.py: Implementation of OnlineBatchMixture algorithm
• skpro/regression/online/_bootstrap_remember.py: Implementation of OnlineBootstrapRemember algorithm

Modified Files

• skpro/regression/online/__init__.py: Added exports for OnlineBatchMixture and OnlineBootstrapRemember

Implementation Details

OnlineBatchMixture

• Fits separate regressor copies on each batch of data
• Returns Mixture distribution with weights proportional to number of samples in each batch
• Supports min_batch_size parameter to control minimum batch size
• Supports ignore_small_batches parameter:
  • If True: Batches smaller than min_batch_size are ignored
  • If False: Small batches are accumulated until min_batch_size is reached

OnlineBootstrapRemember

• Remembers a bootstrap sample of size n_remember from fit data
• At each update, pools remembered sample with new batch data
• Bootstraps a new remembered sample from pooled data to maintain size n_remember
• Returns Mixture distribution combining:
  • Predictions from regressor on remembered sample only
  • Predictions from regressor on pooled data (remembered + new batch)
• Weights are proportional to sample sizes (remembered vs new batch)

Example Usage

from skpro.regression.online import OnlineBatchMixture, OnlineBootstrapRemember
from skpro.regression.residual import ResidualDouble
from sklearn.linear_model import LinearRegression

# OnlineBatchMixture
base_reg = ResidualDouble(LinearRegression())
online_batch = OnlineBatchMixture(
    estimator=base_reg,
    min_batch_size=10,
    ignore_small_batches=False
)
online_batch.fit(X_train_batch1, y_train_batch1)
online_batch.update(X_train_batch2, y_train_batch2)
y_pred_proba = online_batch.predict_proba(X_test)  # Returns Mixture distribution

# OnlineBootstrapRemember
online_bootstrap = OnlineBootstrapRemember(
    estimator=base_reg,
    n_remember=100,
    random_state=42
)
online_bootstrap.fit(X_train, y_train)
online_bootstrap.update(X_new_batch, y_new_batch)
y_pred_proba = online_bootstrap.predict_proba(X_test)  # Returns Mixture distribution