[BUG] Generated reports hard-code version 0.1.0 while CLI is 0.4.9

[mioimotoai-lgtm]

Describe the Bug

A JSON report generated by OpenAgent Eval 0.4.9 identifies itself as version 0.1.0 in metadata.version.

The CLI correctly prints OpenAgent Eval v0.4.9, so a downstream archive or comparison tool sees inconsistent producer versions for the same run.

To Reproduce

At revision c8b9452faa27d7ca0a13ce06d311d9bd0fc4a7ee:

1. Run uv sync --group dev.
2. Create a two-row JSON dataset with question, ground_truth, context, and ground_truth_contexts fields.
3. Use the documented offline configuration:

dataset:
  path: data/questions.json
  format: json
llm:
  provider: mock
  model: mock-model
retriever:
  provider: mock
report:
  output: json
  output_dir: reports
parallel: false

4. Run oaeval run config.yaml.
5. Inspect the generated JSON report.

Observed CLI output:

OpenAgent Eval v0.4.9
Items: 2 | Errors: 0

Observed report metadata:

{
  "metadata": {
    "engine": "openagent-eval",
    "version": "0.1.0"
  }
}

I also ran the focused offline integration test:

pytest tests/integration/test_pipeline/test_mock_e2e.py -o 'addopts=' -q
3 passed in 0.03s

Expected Behavior

Generated reports should use the installed package version (0.4.9 here), ideally from the same single version source used by the CLI.

Source Location

openagent_eval/core/engine.py currently sets the report metadata value directly:

metadata={
    "version": "0.1.0",
    "engine": "openagent-eval",
    ...
}

openagent_eval/__init__.py already exposes __version__ = "0.4.9".

Environment

• macOS 26.5.2, arm64
• Python 3.13.13
• OpenAgent Eval 0.4.9
• Source install via uv sync --group dev