Propose a redesigned migration subsystem that addresses the critical
gaps in MIGRATION.md: C-only API, no transport, server-only authoring,
no adoption path for existing databases, rename-column history loss,
and no recovery from out-of-band schema drift.

Key design shifts: SQL-first surface with no public C API, native DDL
with optional manual cross-dialect override, server-authoritative
linear chain keyed by schema_hash (from_hash → to_hash), privileged
local-to-cloud push with automatic client-side catchup, and a defined
chain-reset recovery path when a developer bypasses the alter
lifecycle.