fix: hide compact models from public listings

[znhisme]

Fix report: hide -openai-compact internal variants and inherit base billing

Summary

This patch keeps *-openai-compact available for internal Codex /v1/responses/compact routing, but prevents those internal variants from being exposed as user-selectable/public pricing models.

It also changes compact billing lookup so a compact variant inherits the base model's configured billing first. For example:

• gpt-5.4-openai-compact now inherits configured billing from gpt-5.4.
• If the base model has no configured billing, existing *-openai-compact wildcard/default fallback behavior remains available.

Why

In suffix-based compact routing builds, deleting compact models from channel abilities can break Codex compaction with errors such as No available channel for model ...-openai-compact.

However, exposing compact variants in public model/pricing endpoints is confusing and can show inaccurate prices. When ModelRatio has gpt-5.4 but does not have gpt-5.4-openai-compact, the compact variant can fall back to the built-in default ratio (37.5) instead of the base model's configured ratio.

Changes

• Filter compact suffix models from public model list responses:
  • token-limited /v1/models path
  • group-enabled /v1/models path
  • channel model list response
  • dashboard model list response
  • enabled model list response
• Filter compact suffix models from /api/pricing output.
• Add CompactBaseModelName helper.
• Update billing lookups so compact variants inherit base model values before wildcard/default fallback:
  • model price
  • model ratio
  • completion ratio / completion ratio info
  • cache ratio
  • create-cache ratio
• Add regression tests for pricing filtering and compact billing inheritance.

Verification

Ran locally:

go test ./setting/ratio_setting ./controller -count=1

Result:

ok  	github.com/QuantumNous/new-api/setting/ratio_setting
ok  	github.com/QuantumNous/new-api/controller

Related upstream context

• QuantumNous/new-api PR fix(openai): remove compact model suffix QuantumNous/new-api#3403: remove compact model suffix, longer-term upstream direction.
• QuantumNous/new-api issue 模型价格无法设置 QuantumNous/new-api#3487: compact special models and pricing configuration.
• QuantumNous/new-api PR fix: /v1/responses/compact default billing QuantumNous/new-api#2766: compact default billing fallback behavior.
• QuantumNous/new-api issues Codex 调用 /v1/responses/compact 时返回 Unknown parameter: 'client_metadata' QuantumNous/new-api#4461/Codex requests to /v1/responses/compact fail with Unknown parameter: 'client_metadata' QuantumNous/new-api#4462: Codex compact routing failures when compact variants are unavailable.
• QuantumNous/new-api PR fix: exclude unconfigured models from pricing plaza QuantumNous/new-api#4371: precedent for filtering unconfigured models from pricing plaza.