v1.10.0 - 5 complexity indicators + 3 new rules

Highlights

• 5 complexity indicators in core/indicators.py: column priors (identity + corruption scoring), sparse-match estimation, full-population matchkey hits, cross-blocking overlap, corruption score. Cheap two run eagerly; expensive three run lazily via IndicatorContext memoization.
• IndicatorContext threads indicators through the policy/rule chain. RefitPolicy.propose gains optional ctx kwarg; controller introspects custom-policy signatures via inspect.signature for backward compat.
• 3 new rules: rule_corruption_normalize adds normalize when blocking column has corruption + identity prior; rule_cross_blocking_disagreement proposes multi-pass when overlap is low; rule_sparse_match_expand lowers threshold + side-channel signal when sample is sparse.
• GOLDENMATCH_AUTOCONFIG_INDICATOR_BUDGET=fast env var gates the two expensive indicators for users who prefer v1.9 wall-clock.

Benchmarks (zero-config, no LLM)

Dataset	v1.9.0	v1.10.0
DBLP-ACM	0.9641	0.9641
Febrl3	0.9443	0.9443
NCVR	0.9719	0.9719
DQbench composite	62.87	66.91
T1 F1	89.3%	88.9%
T2 F1	58.7%	69.0% (+10.3pp)
T3 F1	53.8%	53.8%

Composite gain came from T2 recovery via rule_corruption_normalize and rule_cross_blocking_disagreement.

Honest framing

Shipped at fallback target (≥65), not the spec's primary target (≥70). T2 carried the gain. T3 still at 53.8% — needs richer matchkey design (queued for v1.11). T1 slight regression (-0.4pp) within measurement noise.

Breaking changes

None. Adds optional fields with default-None for v1.9 cache compat. RefitPolicy.propose gains optional ctx kwarg (custom 3-arg policies still work via signature introspection).

Full diff: v1.9.0...v1.10.0