INSERT OR IGNORE idempotency shift with WAL mode on Raspberry Pi SQLite

On Raspberry Pi devices running SQLite, applications that retry failed INSERT statements must reconcile idempotency across schema constraints and journal mode. The goal is to determine whether INSERT OR IGNORE alone, when a UNIQUE constraint is present, safely prevents duplicate rows during retries, or whether WAL mode checkpoint timing introduces visibility gaps that compromise retry reliability. Without a UNIQUE constraint, INSERT OR IGNORE behaves as a standard insert, risking duplication. With WAL active, concurrent readers may observe partially committed rows before checkpoint, affecting observable retry outcomes. This design question hinges on the interaction between conflict resolution semantics and journal mode behavior in ARMv7 and ARMv8 deployments. What is the observable difference in duplicate row frequency between INSERT OR IGNORE and ON CONFLICT DO UPDATE under WAL mode? How does checkpoint interval affect idempotency perception during rapid retries? Can a schema without a UNIQUE column ever be made retry-safe with INSERT OR IGNORE?