Escaping JSON data is incorrect for UPDATEs

Tortoise 0.16.10, Postgresql 12, JSONB column created with fields.JSONField().

When inserting data initially with model.create, the prepared statement passes the data correctly:

INSERT INTO "usermodel" (...) VALUES (...): 
...
 '{"attributes":{"cn":["William Bush"],"createTimestamp":["2020-04-30 16:20:02+00:00"],"creatorsName":["cn=admin,dc=example,dc=org"],"description":["Lt. William Bush"],
...

This inserts a dict with a bunch of data, as expected.

Updating the data generates incorrectly escaped SQL whether using QuerySet.update or model.save.

Queryset:

UPDATE "usermodel" SET "ldap"='"{\n    \"attributes\": {\n        \"cn\": [\n            \"William Bush\"\n        ],\n        \"createTimestamp\": [\n            \"2020-04-30 16:20:02+00:00\"\n        ],\n   

...

Model:

UPDATE "usermodel" SET "ldap"=$1 WHERE "ident"=$2: ['"{\\n    \\"attributes\\": {\\n        \\"cn\\": [\\n            \\"William Bush\\"\\n        ],\\n        \\"createTimestamp\\": [\\n            \\"2020-04-30 16:20:02+00:00\\"\\n        ],\\n        \\"creatorsName\\": [\\n            \\"cn=admin,dc=example,dc=org\\"\\n        ],

...

As you can see both update statements are wrong: instead of inserting a JSON document, they end up inserting a string. Since a string is a valid JSON document, the update is successful but when attempting to read the data later it's trash.

In the Queryset case since it's not using a prepared statement I guess it has to attempt to escape the string. Is it possible to change it to use prepared statements?

In the model case since it's using a prepared statement there is no reason to quote the data; it should be passed untouched like INSERT does.

Reproduce the issue with a PostgreSQL JSONB field created through fields.JSONField(), comparing model.create with QuerySet.update and model.save. Trace how each update path serializes the value and verify that the completed update stores a JSON object that reads back as the original data rather than a JSON string.