bug: a repository model is not treated as an own model when quota is checked #732

Written-by: HugoXK

Expected behaviour

A model selected for a repository is the model the job runs on, so it must count as an own model everywhere: the first-run checklist, the dashboard's quota next action, and the job enqueue quota check. A user with such a model can start a job after the platform quota is exhausted, and the model step is not marked done just because a provider row exists.

When the platform quota is the fallback, the model step must also tell the truth about that quota:

• Show the current remaining platform balance.
• Below PLATFORM_QUOTA_COMFORT_CNY (50 CNY), warn that the balance may not cover the first run and recommend configuring an own model now — or selecting one for the repository when a provider already exists.
• At 50 CNY or more (or when the account is unlimited), show only the remaining balance and keep Settings → Model as the place to configure a model later, with no warning.
• The threshold changes the advice only; it never changes quota accounting or the gateway.

Actual behaviour

• src/app.ts marks the model step done when any provider has any model, even if no repository selected one.
• src/gateway/gateway.ts only uses a repository model when it is in repo_model_selection; otherwise the call is platform-funded.
• src/jobs/service.ts decides platformFunded from the legacy users.own_llm_* fields only, so a repository-selected model still gets QUOTA_EXHAUSTED when the platform quota is empty.
• src/views/pages.ts shows the quota "Add your own model" action unconditionally when the quota is exhausted, even when a usable own model exists.
• The platform-satisfied model step says only "Using platform quota (default)" and never says how much is left or whether that is enough for a first run.

Reproduction

1. Create a user with quota_cny = 0 (or spend the default quota).
2. Add a provider and a model in Settings, then select that model for a repository.
3. Start a job.

Expected: the job is dispatched through the repository model. Actual: enqueue returns QUOTA_EXHAUSTED before the gateway is called. The dashboard also offers "Add your own model" and may mark the model step done before the repository selection exists. A new account with the default 1 CNY quota is treated as if the platform balance were a comfortable default instead of a warning.

Scope

• src/app.ts: derive modelConfigured from a usable own model (legacy own model or a repository selection), keep providerConfigured separate, and pass the remaining platform quota into the checklist.
• src/views/checklist.ts: a provider without a repository selection is not a completed model step; link it to the repository picker. A selected model satisfies the step. When the platform quota is the fallback, show the balance and the low/high advice described above.
• src/views/pages.ts and public/app.css: show the quota warning in the checklist row and suppress the exhausted-quota next action when a usable own model exists.
• src/jobs/service.ts: treat a repository selection as non-platform-funded in both enqueue and dispatch.
• Tests in test/setupChecklist.test.ts, test/nextAction.test.ts, test/jobs.test.ts; docs in docs/architecture.md and docs/quota.md; one changelog fragment.

Out of scope: changing the gateway fallback, quota accounting, or auto-binding a provider to every repository.

Verification

• A repository with a selected model and an exhausted quota enqueues.
• A repository without a selection and an exhausted quota is still refused as today.
• The first-run model step is not done when only a provider exists, and links to the repository picker.
• The quota next action is absent when a usable own model exists.
• A 1 CNY platform balance shows a warning and an "add your own model now" action; a 50 CNY balance (or unlimited) shows only the balance and the later Settings → Model path.
• npm run check passes.