chore(e2e): refresh Cypress fixtures so MiniMax M3 appears in dropdown (#458)

* chore(e2e): refresh Cypress fixtures so MiniMax M3 appears in dropdown

The E2E_FIXTURES=1 cypress build serves availability.json (and the rest of
packages/app/cypress/fixtures/api/) instead of querying the DB. After #456
added MiniMax M3 to the dashboard, the model-architecture.cy.ts spec gained
a new MiniMax M3 describe block that tries to click `MiniMax M3` in the
model dropdown — but the fixture availability.json predated M3, so
GlobalFilterContext.availableModels filtered MiniMax M3 out of the dropdown
and the `cy.contains('MiniMax M3').click()` in the `before all` hook timed
out. The leftover open Radix popper then took out the gpt-oss and DeepSeek
V4 Pro suites with `[role="combobox"]` not visible.

Refreshed via `pnpm --filter app capture:fixtures` (captures from
https://inferencex.semianalysis.com). Verified locally:
`E2E_FIXTURES=1 pnpm start` + `cypress run --spec model-architecture.cy.ts`
→ 45/45 passing.

Co-authored-by: Alec Ibarra <adibarra@users.noreply.github.com>

* fix(e2e): pair i_prec=fp8 URL with a model that has FP8 in fixtures

After PR #435 made DeepSeek-V4-Pro the default model, the refreshed
availability fixture (which includes only FP4 for dsv4) caused
`effectivePrecisions` to intersect the URL-seeded ['fp8'] with the
available set and fall back to FP4. The precision-multiselect then
rendered FP4 instead of the expected FP8.

Add `g_model=DeepSeek-R1-0528` to the two affected specs so the URL
selection is meaningful — matching the workaround already used by the
"multiple URL params" test in url-params.cy.ts.

Co-authored-by: Alec Ibarra <adibarra@users.noreply.github.com>

* chore(fixtures): apply oxfmt formatting to refreshed Cypress fixtures

Co-authored-by: Alec Ibarra <adibarra@users.noreply.github.com>

---------

Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: Alec Ibarra <adibarra@users.noreply.github.com>

Author: adibarra

packages/app/cypress/e2e/line-labels.cy.ts

@@ -80,7 +80,11 @@ describe('Line Labels Toggle', () => {
   });
 
   it('appends the precision to each line label when multiple precisions are selected', () => {
-    cy.visit('/inference?i_linelabel=1&i_prec=fp4,fp8', {
+    // Pair the FP4+FP8 selection with a model that has both precisions in the
+    // fixtures. The default model (DeepSeek-V4-Pro) only has FP4, so
+    // `effectivePrecisions` would drop FP8 and the test couldn't observe the
+    // dual-precision labeling behavior.
+    cy.visit('/inference?g_model=DeepSeek-R1-0528&i_linelabel=1&i_prec=fp4,fp8', {
       onBeforeLoad(win) {
         win.localStorage.setItem('inferencex-star-modal-dismissed', String(Date.now()));
       },

packages/app/cypress/e2e/url-params.cy.ts

@@ -186,7 +186,11 @@ describe('URL Parameter Persistence', () => {
     });
 
     it('/inference?i_prec=fp8 seeds the precision without a hydration error', () => {
-      visitWithErrorSpy('/inference?i_prec=fp8');
+      // Pair `i_prec=fp8` with a model that actually has FP8 in availability.
+      // The default model (DeepSeek-V4-Pro) is FP4-only in the test fixtures,
+      // so `effectivePrecisions` would otherwise intersect the URL selection
+      // with the available set and fall back to FP4.
+      visitWithErrorSpy('/inference?g_model=DeepSeek-R1-0528&i_prec=fp8');
       cy.get('[data-testid="precision-multiselect"]').should('contain.text', 'FP8');
       assertNoHydrationMismatch();
     });