fix(providers): allow replacing preferred models

[penso]

Summary

• Preselect saved provider model preferences when opening the preferred-model dialog.
• Replace a provider's previous preferred models on save, including clearing all preferences with an empty selection.
• Add backend and Playwright regression coverage for de-preferring models.

Validation

Completed

• cargo fmt --all -- --check
• cargo test -p moltis-provider-setup save_models_
• npm run build
• npm run build:all
• npx tsc --noEmit
• npx playwright test e2e/specs/providers.spec.js

Remaining

• Full repository validation not run.

Manual QA

• Open Settings > Providers.
• Open Preferred Models for a provider with existing preferred models.
• Confirm existing preferences are selected, deselect one, save, and confirm only the remaining selected model stays preferred.
• Deselect all preferred models, save, and confirm no models remain preferred for that provider.

Fixes #1094

[greptile-apps]

Greptile Summary

This PR fixes preferred-model preferences so that saving a new selection fully replaces the old one — both in persistent key-store and in the in-memory cross-provider priority list — and pre-populates the dialog with the current saved selection.

• Backend (credentials.rs): save_models_inner now loads the provider's existing model list before writing, then evicts those old entries from the shared priority list before inserting the new selection. Empty saves (deselect-all) are handled correctly because the removal loop still runs, and the insert loop is simply a no-op.
• Frontend (auth-flow.ts): selectedIds is now seeded by matching the available model list against saved preferences using both a direct ID lookup and a namespace-stripped token comparison, fixing the mismatch between key-store bare names (e.g. gpt-4) and models.list fully-qualified IDs (e.g. openai::gpt-4).
• Tests: Two backend unit tests cover the replace and clear scenarios; a new Playwright spec validates end-to-end that the dialog preselects the right cards and emits the correct providers.save_models payload after a deselection.

Confidence Score: 5/5

Safe to merge — the replace logic is correct, the failure path (save_config error) leaves the priority list untouched, and both the backend unit tests and the Playwright spec exercise the key scenarios.

The three changed files are tightly scoped: the Rust side correctly evicts old models before inserting new ones, the TypeScript side resolves the bare-name vs namespaced-ID mismatch, and the tests validate both behaviours end-to-end. No edge cases were found that the new code mishandles.

No files require special attention.

Important Files Changed

Filename	Overview
crates/provider-setup/src/service/implementation/credentials.rs	Adds previous-model eviction from the priority list before inserting the new selection; load/save ordering is correct, error path leaves priority list untouched, and two unit tests verify both replace and clear cases.
crates/web/ui/src/providers/auth-flow.ts	Pre-selects saved preferences in the multi-model dialog using a dual-lookup strategy (direct ID match + namespace-stripped token match) to handle the bare-name vs fully-qualified-ID mismatch between key-store and models.list.
crates/web/ui/e2e/specs/providers.spec.js	New Playwright test mocks both models.list and providers.available, verifies the dialog pre-selects both saved models, deselects one, and confirms the save RPC carries only the remaining selected model.

Sequence Diagram

sequenceDiagram
    participant UI as auth-flow.ts
    participant WS as WebSocket RPC
    participant SVC as save_models_inner
    participant KS as KeyStore
    participant PL as PriorityList (RwLock)

    UI->>WS: "models.list {}"
    WS-->>UI: "[{id:"openai::gpt-a", ...}, ...]"
    UI->>WS: "providers.available {}"
    WS-->>UI: "[{name:"openai", models:["gpt-a","gpt-b"]}]"
    Note over UI: Build selectedIds by matching model.id vs savedModels using stripModelNamespace fallback
    UI->>UI: User deselects gpt-b
    UI->>WS: "providers.save_models {provider:"openai", models:["openai::gpt-a"]}"
    WS->>SVC: save_models_inner(params)
    SVC->>KS: load_config("openai") → ["gpt-a","gpt-b"]
    SVC->>KS: "save_config("openai", models=["gpt-a"])"
    SVC->>PL: write lock acquired
    loop for each previous model
        PL->>PL: "retain(existing != previous)"
    end
    loop for each new model (rev)
        PL->>PL: retain + insert(0, m)
    end
    SVC-->>UI: "{ok: true}"

Loading

_{Reviews (1): Last reviewed commit: "fix(providers): allow replacing preferre..." | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

✅ 39 untouched benchmarks
⏩ 5 skipped benchmarks¹

---

_{Comparing troubled-danger (e8a1b5f) with main (a568a3c)}

Footnotes

1. 5 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[codecov]

Codecov Report

❌ Patch coverage is 98.71795% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...er-setup/src/service/implementation/credentials.rs	98.71%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!