Separate payload key from collection path; remove caller-side pluralization so Document reuses shared CRUD

[TimEvci]

Summary

The singular endpoint string is doing two unrelated jobs: it is the JSON payload wrapper key ({"sales_invoice": {...}}) and the stem of the URL path — with a trailing "s" appended by hand at 24 call sites (f"{self.endpoint}s/..."). This conflation is the root cause of several problems: Document cannot inherit the shared CRUD mixins and reimplements load/find_by_id/delete/save by hand, Synchronizable._sync_endpoint duplicates the snake_case derivation and reaches into cls.__private_attributes__, and every new call site must remember the pluralize-at-call-site convention.

Separating the two concepts deletes all of this rather than centralizing it.

Details

Current state:

• MoneybirdModel.endpoint property (moneysnake/model.py:22-31) derives a singular name; all callers append "s" themselves (24 occurrences across model.py, sales_invoice.py, external_sales_invoice.py, financial_statement.py).
• Synchronizable._sync_endpoint (moneysnake/model.py:130-138) re-derives the same name as a classmethod, pulling the _endpoint default out of cls.__private_attributes__ — a private pydantic registry — because the property needs an instance.
• Document (moneysnake/document.py:91-129) needs the path documents/{resource}s, which the single-string model cannot express, so it reimplements load, find_by_id, delete, and the URL half of save instead of inheriting Loadable/Saveable/Deletable.

Suggested fix

Split the two concepts into two classmethods:

@classmethod
def _payload_key(cls) -> str:
    """JSON wrapper key for request bodies, e.g. "sales_invoice"."""
    # single snake_case derivation, honoring an explicit override

@classmethod
def _collection_path(cls) -> str:
    """URL path of the collection, e.g. "sales_invoices"."""
    return f"{cls._payload_key()}s"

• Document overrides only _collection_path to return f"documents/{cls._payload_key()}s" and keeps _payload_key as the resource name — the payload key and the URL path are different values for documents, which is exactly why they must be separate methods.
• All 24 f"{self.endpoint}s..." call sites become f"{self._collection_path()}/..." (or a small _item_path(id) helper); the hand-appended "s" convention disappears entirely.
• Synchronizable._sync_endpoint and the __private_attributes__ access are deleted; sync_list/sync_fetch use _collection_path.
• Document.load / find_by_id / delete and its _base_path/_sync_endpoint overrides are deleted; it inherits the mixins like every other model.
• The instance endpoint property can delegate to _payload_key() (or be removed if nothing external depends on it).

Related cleanups in the same file (fold in)

• Loadable.load(self, id) takes an id parameter despite being an instance method — SalesInvoice(id=5).load(7) is legal and leaves the object with id=7. load() should take no argument and use self.id (raising if unset); find_by_id currently passes the id twice.
• Deletable.delete_by_id returns the deleted entity with id=None — a husk object with no meaningful use. It should return None.

Both are small public-API changes; note them in the changelog.

Acceptance criteria

• The snake_case derivation exists once; no access to pydantic private-attribute internals remains.
• No call site appends "s" to an endpoint string by hand.
• Document and its subclasses inherit load/find_by_id/save/delete from the shared mixins, overriding only _collection_path (and its payload key via _resource).
• Request URLs and payload wrapper keys are byte-for-byte identical to today's for every entity (verify against the existing tests, which assert full paths).
• Existing tests pass, updated only where they exercise the removed load(id) parameter or delete_by_id return value.