It is not possible. The data models and underlying technologies are different.
Use case for Paratext: If you want to break some words into morphemes but you do not want to rigorously define each morpheme and all of its allomorphs.
Use case for Fieldworks parser: I do want to break all words into morphemes and rigorously define each morpheme and all of their allomorphs.
What the two parsers are good at is different. Moving back and forth between parsers is not advisable.
I would say that if you copy the baseline text from Flex to an extra book in Paratext, it will very quickly start recognizing morphemes if you manually divide words with the most common morphemes. You can give it a try and see if you like the results.
My suggestion for teaching the Paratext Parser quickly is to use the wordlist tool and not the project interlinearizer to find and parse some of the most productive words. For example sort the word list tool for a very common verb and manually parse several surface forms of that verb. The Paratext parser will take that information and will apply that to other verbs. It will quickly learn to recognize the tense/aspect markers for the language.