test: remove seasonal xfail and tighten R equivalence tolerances

State-space seasonal smoother achieves <0.1% relative error vs R bsts,
far exceeding the original ±5% tolerance with dummy-variable approach.

Tolerance changes:
- seasonal: ±5% → ±1% (actual: <0.1%)
- basic/strong_effect CI: ±1.5% → ±1.0% (actual: <0.5%)
- Remove xfail from TestEquivalenceSeasonal (all 5 tests pass)

All 33 numerical equivalence tests pass with tightened tolerances.

Author: YuminosukeSato

tests/test_numerical_equivalence.py

@@ -43,13 +43,14 @@
 }
 
 TOL_POINT = 0.03  # ±3% for point estimates
-# ±1.5% for no-covariates CI bounds
-# (convergence ~0.83%, MCMC variance ±0.5%)
-TOL_CI_NO_COV = 0.015
+# ±1.0% for no-covariates CI bounds
+# (actual error ~0.44%, MCMC variance margin included)
+TOL_CI_NO_COV = 0.01
 TOL_CI_COV = 0.01  # ±1% after aligning with the R static regression prior
 TOL_CI_PHASE2 = 0.03  # explicit Phase 2 benchmark for covariate CI bounds
-TOL_POINT_SEASONAL = 0.05
-TOL_CI_SEASONAL = 0.05
+# State-space seasonal (DK simulation smoother): actual error < 0.1%
+TOL_POINT_SEASONAL = 0.01
+TOL_CI_SEASONAL = 0.01
 ABS_TOL_NO_EFFECT = 2.0  # absolute tolerance when true_effect=0
 
 
@@ -300,10 +301,6 @@ def test_p_value_significance(self):
         _assert_significance(py, fixture["r_output"])
 
 
-@pytest.mark.xfail(
-    reason="seasonal model migrating from static regressors to state-space (R bsts compat)",
-    strict=False,
-)
 class TestEquivalenceSeasonal:
     SCENARIO = "seasonal"