Decoupling TSV metadata schema parsing from API endpoint

Currently, any TSV file loaded into a Dataverse instance flows through DatasetFieldServiceApi.loadDatasetFields().
This method isn't just the endpoint, it also contains all the code to parse and convert the TSV into living model objects.

This needs to be torn apart:

• Enable proper testing of the parser
• Enable reading TSV metadata schemas during unit & integration testing. A bare necessity for #5989
• Enable reading the files not only via API endpoint:
  • Possibility to move the code into a reusable module outside main codebase
  • Load TSVs from disk/WAR during startup and bootstrap (Hello, #5361)
• Enable validation on load
• Refactor some dusted code