Processing of a file should be based on TMs set up in DB, not provided in a processing request

[ArchiDevil]

Currently the file processing request requires setting up of TM IDs:

class DocumentProcessingSettings(BaseModel):
    substitute_numbers: bool
    machine_translation_settings: Optional[MachineTranslationSettings]
    memory_ids: list[int] # HERE IS 
    memory_usage: TranslationMemoryUsage
    similarity_threshold: float = Field(default=1.0, ge=0.0, le=1.0)

This is quite weird, because we have a special endpoint to set up things for a document and this request duplicates the code. Even worse, this code on backend does not bind TMs to a document and user is required to set up TMs afterward manually. This must be done using the same endpoint and worker should use IDs from the database, not ones passed within a task.