fix: escaped Unicode in BaseAdapter

[leonardmq]

What does this PR do?

Fix a bug where Unicode in structured input and output is escaped in the BaseAdapter, causing two main issues:

1. Structured input and output are sent back to the frontend as escaped strings (code reference), which are displayed as-is in the Datasets page.
2. Structured input and output are persisted in .kiln files as ASCII. While valid JSON, this is not ideal for scenarios like collaboration, committing datasets to GitHub, and searching. Additionally, there is a risk of improperly decoding these strings.

After this fix, Kiln will store structured input and output without escaping Unicode, which is already the behavior for plain text input and output. This change is non-breaking, as json.load and json.loads can handle both Unicode and ASCII-escaped Unicode.

Related Issues

Part of a group of bugs caused by escaped Unicode: #95

Contributor License Agreement

I, @leonardmq, confirm that I have read and agree to the Contributors License Agreement.

Checklists

• Tests have been run locally and passed
• New tests have been added to any work in /lib

[leonardmq]

Run summary output preview before / after:

[leonardmq]

Task before (with escaping):

{
  "v": 1,
  "id": "179446515248",
  "created_at": "2025-01-13T22:36:58.663778",
  "created_by": "leonardmarcq",
  "input": "I like cars.",
  "input_source": {
    "type": "human",
    "properties": {
      "created_by": "leonardmarcq"
    }
  },
  "output": {
    "v": 1,
    "id": "172557301377",
    "path": null,
    "created_at": "2025-01-13T22:36:58.663734",
    "created_by": "leonardmarcq",
    "output": "{\"translation\": \"\\u6211\\u559c\\u6b22\\u6c7d\\u8f66\\u3002\", \"sentiment\": 4}",
    "source": {
      "type": "synthetic",
      "properties": {
        "adapter_name": "kiln_langchain_adapter",
        "model_name": "gpt_4o",
        "model_provider": "openrouter",
        "prompt_builder_name": "simple_prompt_builder"
      }
    },
    "rating": null,
    "model_type": "task_output"
  },
  "repair_instructions": null,
  "repaired_output": null,
  "intermediate_outputs": {},
  "tags": [
    "manual_run"
  ],
  "model_type": "task_run"
}

Task after (without escaping):

{
  "v": 1,
  "id": "305769926316",
  "created_at": "2025-01-13T22:37:19.291991",
  "created_by": "leonardmarcq",
  "input": "I like cars.",
  "input_source": {
    "type": "human",
    "properties": {
      "created_by": "leonardmarcq"
    }
  },
  "output": {
    "v": 1,
    "id": "272336414910",
    "path": null,
    "created_at": "2025-01-13T22:37:19.291943",
    "created_by": "leonardmarcq",
    "output": "{\"translation\": \"我喜欢汽车。\", \"sentiment\": 4}",
    "source": {
      "type": "synthetic",
      "properties": {
        "adapter_name": "kiln_langchain_adapter",
        "model_name": "gpt_4o",
        "model_provider": "openrouter",
        "prompt_builder_name": "simple_prompt_builder"
      }
    },
    "rating": null,
    "model_type": "task_output"
  },
  "repair_instructions": null,
  "repaired_output": null,
  "intermediate_outputs": {},
  "tags": [
    "manual_run"
  ],
  "model_type": "task_run"
}

[scosman]

Thanks @leonardmq !