docs(dea): Improve Readme file to be skill compatible

[minzznguyen]

Desciption

This PR makes changes to README.md file so that when a new user tells an AI Agent to read the README, it can start running evalbench evaluations with a small amount of guidance

For example, this can be a prompt of a user to the agent:

can you read this file -  evalbench/datasets/dea-tools/README.md
and run an eval case for me:
- User: read the schema of this public BigQuery table samples.shakespeare
- Expected Agent Result:
BigQuery Public Dataset: samples.shakespeare 
Table Schema: 
- word (STRING): A single unique word extracted from a corpus. 
- word_count (INTEGER): The number of times this word appears in the corpus. 
- corpus (STRING): The specific work (play/poem) from which the word was extracted. 
- corpus_date (INTEGER): The year in which the corpus was published.
create all new files under evalbench/sandbox

Agent generated the right YAML and dataset JSON file:

orchestrator: dea
dataset_format: dea-format
model_config: datasets/model_configs/gcp_data_engineering_agent_model.yaml
simulated_user_model_config: datasets/model_configs/gemini_2.5_pro_model.yaml
dataset_config: sandbox/shakespeare.evalset.json
env:
  SCENARIO_ID: "dea-shakespeare-schema"
scorers:
  binary_rubric_scorer:
    model_config: datasets/model_configs/gemini_2.5_pro_model.yaml
  dataform_cloud_compile:
    gcp_project_id: !ENV ${EVAL_GCP_PROJECT_ID}
    gcp_region: !ENV ${EVAL_GCP_PROJECT_REGION}
    timeout_seconds: 300
  dataform_cloud_run:
    gcp_project_id: !ENV ${EVAL_GCP_PROJECT_ID}
    gcp_region: !ENV ${EVAL_GCP_PROJECT_REGION}
    timeout_seconds: 300
reporting:
  csv:
    output_directory: results
  bigquery:
    gcp_project_id: !ENV ${EVAL_GCP_PROJECT_ID}
runners:
  agent_runners: 1
set_up_script: datasets/dea-tools/scripts/setup_dataform.sh
tear_down_script: datasets/dea-tools/scripts/teardown_dataform.sh
dataform_workspace_gcs_archive:
  bucket: !ENV ${EVAL_GCP_PROJECT_ID}-evalbench-archives
  path_prefix: workspaces
  gcp_project_id: !ENV ${EVAL_GCP_PROJECT_ID}
  gcp_region: !ENV ${EVAL_GCP_PROJECT_REGION}

And JSON file

{
  "scenarios": [
    {
      "id": "shakespeare-schema",
      "starting_prompt": "read the schema of this public BigQuery table samples.shakespeare",
      "max_turns": 1,
      "conversation_plan": [],
      "binary_rubric": [
        "Verify that the agent listed the table schema for samples.shakespeare with columns: word (STRING), word_count (INTEGER), corpus (STRING), corpus_date (INTEGER) and their descriptions, matching: - word (STRING): A single unique word extracted from a corpus. - word_count (INTEGER): The number of times this word appears in the corpus. - corpus (STRING): The specific work (play/poem) from which the word was extracted. - corpus_date (INTEGER): The year in which the corpus was published."
      ]
    }
  ]
}

The agent ran the correct command and returned back the result of the eval case as PASS