feat: dynamic max_iter and tol for MICE via seven-signal framework

[DEVunderdog]

Parent

#91

What to build

Replace the hardcoded max_iter=10 and default tol in the MICE IterativeImputer with values computed from a seven-signal framework, aggregated conservatively across all MICE columns. Add post-fit convergence monitoring: if n_iter_ == max_iter, append a convergence warning to every MICE column's ColumnImputationRecord.signals.

The seven signals and their MICE-specific aggregation rules:

1. NonlinearityTag — most complex tag across MICE columns; ComplexNonlinear adds to base max_iter
2. Feature matrix missingness fraction — fraction of NaN cells across the full MICE matrix
3. R² strength — minimum R²_linear across MICE columns (worst-case convergence speed); high minimum R² reduces max_iter
4. Inter-feature correlation — maximum pairwise Pearson |r| among MICE columns (from CorrelationProfiler); high correlation increases max_iter
5. Complete row fraction — fraction of rows with no NaN across all MICE columns; low fraction increases max_iter
6. Scale-relative tol — tol = min_iqr * scaling_factor where min_iqr is the minimum IQR across MICE columns
7. ComplexNonlinear tol tightening — tighter tol when most-complex tag is ComplexNonlinear

Implement as _compute_mice_max_iter and _compute_mice_tol helpers, parallel to the existing _compute_max_iter / _compute_tol used by the Regression strategy.

Acceptance criteria

• IterativeImputer for the MICE block is constructed with dynamically computed max_iter and tol (not hardcoded 10 / sklearn default)
• _compute_mice_max_iter applies all seven signals with MICE-specific aggregation (min R², max inter-column |r|, full-matrix missingness)
• _compute_mice_tol uses min_iqr across MICE columns, with tighter scaling for ComplexNonlinear
• Post-fit: if n_iter_ == max_iter, a convergence warning is appended to every MICE column's signals (not only the first or last)
• Unit test: a MICE block engineered to hit max_iter before convergence produces a convergence warning signal on all MICE column records
• Unit test: a well-conditioned MICE block that converges early does not produce a convergence warning
• Numpy-style docstrings on _compute_mice_max_iter and _compute_mice_tol

Blocked by

• Blocked by feat: MICE estimator selection via NonlinearityTag + Unpredictable all-fallback #158