fix(dag): synthesize final-gate binding and review gate error wording - #306

Closes #304 and #305 — both discovered by live end-to-end testing of v1.0.21.

#304 — false unresolved_reviews after ACCEPT

workflow(action="status") reported unresolved_reviews for a diff review that settled ACCEPT and completed the graph. Root cause: finalReviewGates requires the final gate to map the review's output; the block compiler emitted the follow-up synthesize with no input_mapping, so no gate was ever found and an ACCEPTed review stayed "unresolved" forever. Same contract-vs-binding pattern as the pre-#299 verify gap.

Fix: a synthesize depending on a review now binds that review's output into its input_mapping (its condition is already <review>.output.verdict == "ACCEPT"), satisfying all three finalReviewGates axes. Side effect: block routes shaped … → review → synthesize become valid under deep-mode validateFinalReviewGate too, removing the mode cliff.

Tests: new compile-seam case asserts the binding exists (and is absent without a review dependency); existing unresolvedReviewOutcomes cases already prove the resolver accepts this shape.

#305 — gate message claimed "deep workflow" in all modes

ReviewGateError wording dropped the unsupported mode claim (gate behavior unchanged). The wake-integration test now pins the message free of the "deep workflow" wording.

Verification

• 550 DAG/tool tests green (72 focused), bun typecheck clean.