Build ScenarioRuleEngine evaluate() core algorithm

Implement evaluate() as a method on a ScenarioRuleEngine class that takes PromptHistoryRepository as a constructor-injected dependency. Use the Chain of Responsibility pattern: define a private List where each evaluator checks one condition dimension and returns a ConditionResult. This makes each condition independently testable and the overall algorithm easy to extend. Process conditions in this order: (1) contact type, (2) wellbeing flags, (3) duration range, (4) delay window, (5) cooldown guard — ordering from cheapest to most expensive (the cooldown check requires async I/O and should be last).

EvaluationReasonCode should be an enum with values: MATCHED, NO_RULES_CONFIGURED, NO_RULES_MATCHED, CONTACT_TYPE_MISMATCH, WELLBEING_FLAGS_NOT_MET, DURATION_OUT_OF_RANGE, DELAY_WINDOW_NOT_MET, COOLDOWN_ACTIVE. Store conditionTrace as a List where each entry captures (conditionName, passed, actualValue, expectedValue) for full observability. The PromptHistoryRepository call should use a DateTime parameter rather than DateTime.now() directly — inject a clock abstraction for testability.

Write unit tests using flutter_test with a mocked PromptHistoryRepository. Test matrix must cover: contact type exact match, contact type wildcard, contact type mismatch; wellbeing 'any' semantics with one match, all match, no match; wellbeing 'all' semantics with partial match (should not match) and full match; duration at min boundary (inclusive), max boundary (inclusive), below min, above max; delay window within range, before min, after max; cooldown active (recent prompt exists), cooldown inactive (no recent prompt or expired); empty rules list; all conditions satisfied simultaneously. Use parameterized test cases (test.parametrize pattern) to cover boundary values. Target 100% branch coverage on the evaluate() method body.

Add a performance smoke test asserting evaluate() completes in under 200ms for a list of 50 rules.