feat: Z.AI usage bars in Models settings

[pridemusvaire]

Summary

• Poll Z.AI's monitor quota endpoint (/api/monitor/usage/quota/limit) for connected Z.AI and Z.AI Coding Plan providers
• Render 5-hour / weekly (and monthly tools) usage bars under those rows on Models settings, same surface as Kimi / ChatGPT subscription usage
• Region-aware host selection: International → api.z.ai, China → open.bigmodel.cn

Implementation

Layer	Change
packages/types	zai / zai-coding-plan on SubscriptionUsageProviderId; host + path constants
packages/db	parseZaiQuotaUsage, fetchZaiUsage, fetchZaiCodingPlanUsage; included in getSubscriptionProviderUsage
apps/web	Pass usage into ConnectedProviderRow for both Z.AI providers

Upstream payload (undocumented, same as Coding Plan tools / pi-glm-usage):

{
  "code": 200,
  "data": {
    "level": "lite",
    "limits": [
      { "type": "TOKENS_LIMIT", "unit": 3, "percentage": 16, "nextResetTime": 1777819631597 },
      { "type": "TOKENS_LIMIT", "unit": 6, "percentage": 4, "nextResetTime": 1778262784969 }
    ]
  }
}

Missing credentials or unparseable responses still soft-fail (no bar), matching existing providers.

Test plan

• Unit tests for parseZaiQuotaUsage and both fetchers (global + china hosts, coding-plan key)
• Connect Z.AI with an API key → Models settings shows usage bars when the monitor endpoint returns windows
• Connect Z.AI Coding Plan → same bars under that row
• China region uses open.bigmodel.cn
• No key / 401 / empty limits → row stays connected with no usage line (no error toast)

[roomote-roomote]

No code issues found. See task

_{Reviewed b7ffb65}