fixes https://github.com/cloudquery/cloudquery-issues/issues/2763
(internal issue)

Normally, the JSON Flattener will attempt to flatten JSON objects to the first level, e.g.:

"type_schema": "[{\"availability_domain\":\"utf8\",\"display_name\":\"utf8\",\"volume_group_replica_id\":\"utf8\"}]"

But the TypeSchema notation had a special case that the JSON Flattener wasn't aware of:

      {
        "name": "freeform_tags",
        "type": "json",
        "type_schema": "{\"utf8\":\"utf8\"}"
      },

In this case, the utf8 key is not a literal string, but a special value that means "any string key". This should be ignored by the flattener, because it's meant to contain different keys on different rows, and table schemas are fixed.

The only way to implement this would be to do a linear pass over all rows and learn all possible keys, then do another pass and create nullable columns on all non-supplied columns for each row. We definitely don't want to support this.

Instead, this PR simply ignores keys with this special literal string.

The practice is common on the Oracle source plugin; here's a flattened Oracle table that was suffering from this issue, now fixed: