feat: wire altimate-core 0.5.1 equivalence dialect + decidable into dbt review

[anandgupta42]

Description

altimate-core was bumped 0.4.0 → 0.5.1 (#925). 0.5.1 adds net-new equivalence surface that altimate-code does not yet consume:

1. checkEquivalence(sqlA, sqlB, schema, dialect?) — a new optional dialect parameter. The dbt PR reviewer already resolves and threads a dialect end-to-end (orchestrate → runner → dispatcher), but a prior commit (89f77a75ca) had to drop it at the engine boundary because 0.4.0 only accepted 3 params. The carefully-resolved dialect never reaches the engine.
2. EquivalenceResult.decidable: boolean — a new authoritative flag for "the engine could not decide." The runner currently guesses decidability via validation_errors.

Goal

• Forward the dialect hint to core.checkEquivalence in the native equivalence handler and the sql.diff handler so dialect-specific compiled warehouse SQL parses and becomes decidable instead of abstaining on a syntax error.
• Honor the engine's authoritative decidable flag in the review runner (strictly safer — only ever abstains more; preserves FP-safety).
• Add extensive E2E tests against the real engine, including the full runReview pipeline.

Acceptance

• dialect reaches core.checkEquivalence in both handlers
• Empty-string dialect (the ReviewConfig.dialect default) is coerced to "no hint" (engine throws on unknown dialect '')
• Runner honors decidable === false and stays backward-compatible when the field is absent (0.4.0 shape)
• Real-engine native tests + full-pipeline E2E tests, all green; typecheck + marker check clean