[P0] Clarify eval engine support and remove unsupported headless examples

[spazyCZ]

Objective

Align the eval engine contract, examples, and compatibility text so supported and unsupported eval runtimes are described consistently.

Priority

P0 — Must Fix for v1.0

Details

The spec currently defines claude-code and codex as supported eval engines, with copilot and cursor reserved for future use. However, some eval execution-flow examples still show a Copilot headless command. This leaves the supported-engine contract internally inconsistent.

Acceptance Criteria

• Eval examples use only supported engines or clearly mark future placeholders as non-normative
• Unsupported-engine behavior is defined consistently
• The package format and compatibility sections agree on eval engine support

Notes

Source: work/eval-design-discussion.md