Premise: Say we create a lead in Marketo and don't sync it to SFDC. Next, the same lead is created on the SFDC side and synced to Marketo.
Will this create a duplicate lead on the Marketo side since it doesn't have an SFDC ID?
This is why we always recommend not doing this.
As soon as a lead is created in Marketo, push it to SFDC and assign it to a lead queue, until it's ready for sales, inside sales or telemarketing assignment. Lead queues offer a very flexible mechanism in SFDC to store leads and more or less hide them from the sales.
Yes. There's no way for Marketo to know this record represents the same lead (remember, email address is not actually a unique identifier, merely a primary dedupe key for some entry points into the system).
Retrieving data ...