Python: [Bug]: Filtered vector delete tools reject MongoDB ObjectId keys

@eavanvalkenburg is already working on this.

Since Sep 18, 2026.

### Description

`create_delete_tool` fails for a MongoDB collection with `ObjectId` keys when a fixed filter matches a record. The same key works without the filter.

The filtered path serializes the retrieved record through `_encode_vector_tool_record`, which turns its key into a JSON string. It then casts that string to `KeyT` and passes it to `collection.delete`. A cast does not restore the native key, so MongoDB rejects it before sending a delete request.

The expected behavior is to delete the matching records using native `ObjectId` keys and return their JSON strings in `processed_keys`. An empty match should remain a no-op.

### Code Sample

This isolates the problem without a database: database I/O is mocked; the delete tool and MongoDB key validation are real.

```python
import asyncio
from unittest.mock import AsyncMock

from agent_framework import (
Filter,
VectorStoreCollectionDefinition,
VectorStoreField,
create_delete_tool,
)
from agent_framework_mongodb import MongoDBCollection
from bson import ObjectId
from pymongo import AsyncMongoClient

async def main():
key = ObjectId()
client = AsyncMock(spec=AsyncMongoClient)
client.get_database.return_value.get_collection.return_value.delete_many = AsyncMock()
collection = MongoDBCollection(
dict,
definition=VectorStoreCollectionDefinition([
VectorStoreField("key", name="id", type_="ObjectId"),
VectorStoreField("data", name="tenant", type_="str", is_indexed=True),
]),
collection_name="documents",
database_name="vectors",
async_client=client,
)
collection.get = AsyncMock(return_value=[{"id": key, "tenant": "a"}])
delete_tool = create_delete_tool(collection, filter=Filter("tenant", "eq", "a"))
await delete_tool.invoke(arguments={"keys": [str(key)]}, skip_parsing=True)

asyncio.run(main())
```

### Error Messages / Stack Traces

```text
agent_framework.exceptions.IntegrationException: Error deleting records from collection 'documents': MongoDB key values must match the declared 'ObjectId' key type.
```

### Package Versions

- Source: `main` at `0c9944cc9f577d51277ac7c55dbc388b60a577af`
- agent-framework-core: 1.19.0
- agent-framework-mongodb: 1.0.0a260918
- pymongo: 4.18.1

### Python Version

Python 3.12.14 on macOS.

### Additional Context

I have a focused fix using the collection's existing `key_from_json` hook before deletion. The regression test exercises the real MongoDB collection with a mocked driver, including a matching subset, a mapped tenant field, and an empty match. It fails on the current source for the key-type error above; no live MongoDB integration test was run. I will submit the fix with this report.