Hah, i thought of the same a while ago and still like the idea. but with DSP (digital sound processing) adding these types of hooks to a processing loop can become costly very quickly.
Technically i think it is possible. But i also question the feasiblity. How much refactoring would be required and what kind of overhead does that generate performance wise?
While i don’t think the performance hit would be drastic for that itself… the moment you would allow 3rd party devs to create transformers/processors for instruments/steps etc, how do you make sure they don’t hog all the performance with their code? 
On the other hand, if all step effects etc, become “plugins” you’d only have to process/load whatever has been actually used.
It’s all very much dependent on the actual codebase and of course hardware capabilities.