test(e2e): NeuroPage locators reference stale #btn-*-sticky IDs

[alikahwaji]

Summary

tests/pages/NeuroPage.ts:71-74 still targets #btn-start-sticky, #btn-pause-sticky, #btn-step-sticky, #btn-reset-sticky — none of those IDs exist anywhere in src/ or index.html. The current DOM uses #btn-start, #btn-pause, #btn-step, #btn-reset (no -sticky suffix), as defined in index.html:192-203 and src/utils/UIFactory.ts:105-108.

Effect: every E2E spec that touches startButton / pauseButton / stepButton / resetButton through the page object resolves to an empty locator, the await expect(...).toBeVisible() call hangs for 30s, the retry kicks in, and the e2e job burns its full 15-minute timeout on main. Documented in docs/CICD_AUDIT_REPORT.md (commit 240b143).

This is pre-existing drift from the f1e5c2f controllers refactor, not a regression from the recent CI/CD overhaul (0493999).

Repro

```bash
grep -n "btn-.*-sticky" tests/pages/NeuroPage.ts

tests/pages/NeuroPage.ts:71: this.startButton = page.locator('#btn-start-sticky');

tests/pages/NeuroPage.ts:72: this.pauseButton = page.locator('#btn-pause-sticky');

tests/pages/NeuroPage.ts:73: this.stepButton = page.locator('#btn-step-sticky');

tests/pages/NeuroPage.ts:74: this.resetButton = page.locator('#btn-reset-sticky');

grep -rn "btn-.*-sticky" src/ index.html

(no matches — the sticky variant does not exist)

```

Fix

Update tests/pages/NeuroPage.ts:71-74 to drop the -sticky suffix:

```ts
this.startButton = page.locator('#btn-start');
this.pauseButton = page.locator('#btn-pause');
this.stepButton = page.locator('#btn-step');
this.resetButton = page.locator('#btn-reset');
```

After fixing, audit the rest of NeuroPage.ts and each tests/e2e/*.spec.ts for other drift against the post-f1e5c2f DOM. Known-stable IDs (confirmed via grep): #status-accuracy, #status-loss, #status-val-loss, #status-epoch, #viz-container, .dataset-preview-card, #dataset-select.

Acceptance criteria

• tests/pages/NeuroPage.ts uses only IDs that exist in index.html / src/
• npx playwright test runs locally against vite preview and passes or fails on real assertions (not on locator-wait timeouts)
• The e2e job in CI completes well under its 15m budget on main
• No regression — the workflow's continue-on-error: true escape hatch for main stays in place until this is proven green

Priority

Medium. Deploy is already unblocked (via deploy: needs: [test, build]), so this is quality-of-signal work, not an outage. Once it lands, continue-on-error can be removed from the e2e job on main.