Add a --json mode to evals

[sgoedecke]

Adds a --json flag to gh models eval so it can output machine-readable results:

@sgoedecke ➜ /workspaces/gh-models (sgoedecke/evals-json-mode) $ ./gh-models eval --json fixtures/test_single_evaluator.yml
{
  "name": "Test Single Evaluator",
  "description": "Testing a single built-in evaluator",
  "model": "openai/gpt-4o",
  "testResults": [
    {
      "testCase": {
        "expected": "Machine learning is a subset of artificial intelligence that enables computers to learn and make decisions from data without being explicitly programmed.",
        "input": "What is machine learning?"
      },
      "modelResponse": "Machine learning is a subset of **artificial intelligence (AI)**  ...  making it a crucial technology in modern AI-driven applications.",
      "evaluationResults": [
        {
          "evaluatorName": "fluency-test",
          "score": 1,
          "passed": true,
          "details": "LLM evaluation matched choice: '5'"
        }
      ]
    }
  ],
  "summary": {
    "totalTests": 1,
    "passedTests": 1,
    "failedTests": 0,
    "passRate": 100
  }
}

Also adopts some refactors suggested in #54 (extracting some functions and adding a docs link)

[Copilot]

Pull Request Overview

Adds a new machine-readable output mode (via --json) to the gh models eval command and refactors chat‐completion helpers into the prompt package.

• Introduce --json flag to emit structured JSON summary instead of human output.
• Extract GetAzureChatMessageRole and BuildChatCompletionOptions into pkg/prompt/prompt.go for reuse.
• Update CLI handlers, tests, and documentation to support JSON mode.

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
pkg/prompt/prompt.go	Added role‐parsing and option‐building helpers; missing import needed.
cmd/run/run.go	Refactored message‐role switch to use new prompt helper.
cmd/eval/eval.go	Implemented --json flag, JSON serialization, and refactored model call.
cmd/eval/eval_test.go	Added tests covering JSON output format and CLI behavior.
README.md	Documented gh models eval usage, including JSON flag.

Comments suppressed due to low confidence (1)

pkg/prompt/prompt.go:9

• The new GetAzureChatMessageRole function uses fmt.Errorf but fmt is not imported; add import "fmt" to the import block.

github.com/github/gh-models/internal/azuremodels

[johan-j]

Nice job on the json support 🚢

[cheshire137]

Thanks for the new functions!