Fix E2E tests for run-picker/toolbar refactor (#94)

[camronh]

The UI ergonomics overhaul in #94 changed several selectors that the E2E tests depend on:

• .compare-dropdown → .run-picker, .compare-option → .run-picker-row
• Columns/Export controls moved from overflow menu to main toolbar
• alert() → toast notifications
• page.click("body") unreliable for closing panels (hits overlapping elements)

Tests affected:

• test_full_scenario.py::test_full_serve_flow_end_to_end (xfail'd for now)
• test_run_controls.py::test_stop_marks_pending_as_cancelled (flaky timing, xfail'd)
• test_run_controls.py::test_button_shows_stop_when_running (flaky timing, xfail'd)

Partial fixes already applied in 0.1.10:

• test_ui_interactions.py::test_sort_and_toggle_columns - fixed (removed overflow menu open before columns toggle)
• test_ui_interactions.py::test_png_export_modal_allows_configurable_preview - fixed (removed overflow menu open before export toggle)

Remaining work:

• Full scenario test needs comprehensive selector updates for run picker, toolbar, and filter panel close patterns
• Run controls tests need more resilient timing for play→stop transitions