We retain YangParser implementation in ProcessorModuleReactor, as it is populated with ASTSchemaSource representation of local modules. The problem is that we do not ditch the parser after we have assembled SchemaContext in the next step, hence we are retaining also all the scratch objects in BuildGlobalContext.

This leads to sub-optimal performance, as all the BuildGlobalContext memory cannot be freed and used to service actual code generation plugins.

Make sure we ditch the parser as soon as we do not needed it.