Meltano

schema.json

Hi, I was going through the flattening logic in singer-sdk and came across a problem:
the schema (attached below) and a record for that schema (in the thread) are giving error after flattening, on json validation:

```{
    "attributes": {
        
        "lastMessageIn": {
            "id": "64c84b5ad14f695ce4d5160b",
            "sentAt": "2023-08-01T00:01:30.659Z",
            "createdAt": "2023-08-01T00:01:30.914Z",
            "meta": {
                "trackingId": "64c83dbeb0bc8f291035157e",
                "to": null
            },
            "channel": "chat"
        }
    }
}```

I think there's problem in how null is handled in schema flattening vs record flattening

<https://github.com/meltano/sdk/blob/b4f9ac59742300b00e77571156d432ce97df3c05/singer_sdk/helpers/_flattening.py#L238>

if the schema defines that a field can be array or null then the flattened schema should be `{"type": ["string", "null"]}`

Right now I can handle this by using post_process of the stream and replace `null` value to be empty `[]` but I think something is off here

The schema and record are a part from kustomer tap

I can confirm that by adding the following code:

```elif "array" in v["type"] or "object" in v["type"] and max_level &gt; 0:
                if "null" in v["type"]:
                    items.append((new_key, {"type": ["string", "null"]}))
                else:
                    items.append((new_key, {"type": "string"}))```
the stream works.

I've opened an <https://github.com/meltano/sdk/issues/1995|issue>, can immediately send a PR if given a green flag :slightly_smiling_face:

PRs welcome! You may need to update the snapshots and perhaps add <https://github.com/meltano/sdk/blob/b4f9ac59742300b00e77571156d432ce97df3c05/tests/core/test_mapper.py#L638|a new test case> and <https://github.com/meltano/sdk/blob/b4f9ac59742300b00e77571156d432ce97df3c05/tests/core/test_mapper.py#L418-L422|even a new field> with the required=True flag (non-nullable).