feat: provision media billing atomically

[3metaJun]

Description

Adds atomic provisioning for tiered_expr media billing and preserves Grok-compatible image request fields through the OpenAI image DTO. The provision endpoint validates billing mode and expression pairs before replacing channels and options in one transaction. MiniMax keeps its explicit aspect_ratio compatibility.

Type of change

• New feature
• Bug fix

Verification

• go test ./dto ./controller ./model ./relay/channel/minimax ./relay/helper
• go test ./service -run 'TestTryTieredSettle|TestPostTextConsumeQuota' -count=1
• Real local new-api E2E through the apikey.fun Grok upstream:
  • 1K standard image returned one URL and charged the exact generated tier
  • 2K quality request with n=2 returned two URLs and charged the exact generated tier
• Reviewed by gpt-5.6-luna with model_reasoning_effort=max: PASS

Checklist

• The description was reviewed and written for this change.
• I searched existing work and kept this PR focused.
• I understand the implementation and its impact.
• Local tests and a real upstream E2E passed.
• No credentials or unrelated files are included.

---

Agent: gpt-5.6-sol via T3 Code
Review: gpt-5.6-luna (max) via Codex CLI