I'm not sure I agree that an engineer experienced in building scalable systems will need 3 extra weeks to pick a schema and data access pattern that's scalable.
You don't have to actually _use_ spanner or cockroach or vitess in your initial prototype if you don't want to, but it seems entirely reasonable to have an _idea_ of what the story would look like to migrate to a horizontally-scalable approach if product gains traction so it doesn't end up being a multi-year project later.