Most source-centric Dataverse pipelines I see are built by experienced architects from .NET or Java backgrounds.
They are not naive about source control. They know what good ALM looks like on every other platform they have used.
The patterns work in their original context. The transplant does not survive Dataverse.
The platform is structurally environment-centric. You cannot author a Flow as source. You cannot start a model-driven app from XML. The maker portal is where things get built. The unpacked files are downstream.
The pattern is not stupid. The transplant is.
Take what transfers: plugins in C#, web resources in TypeScript, Code Apps in React. Source-centric on every platform including this one.
Reject what does not: Flows, model-driven apps, Canvas Apps, configuration. Environment-centric on Dataverse no matter how many files you commit.
Knowing which is which is the work.