test(ui): end-to-end AppTest tests for all 9 tabs; coverage floor in CI

[Hj1308]

Summary

• New tests/test_app_ui.py (marker ui): drives the real Streamlit app with streamlit.testing.v1.AppTest, with in-memory CSV/XLSX uploads.
  • Smoke: 9 tabs load; CSV and XLSX uploads produce no exception or error in any tab.
  • Tab 1: one summary row per catalyst; best model ∈ MODEL_NAMES; manual exclusion drops "points used" by one; a Time (h) header warns; a t=0 row leaves R² unchanged.
  • Tabs 2–9: each changes its main input and checks a concrete output (linearisation R² table, removal plots, mass-normalised TOF, Arrhenius inputs, H₂O₂ efficiency table, comparison axes, Arrhenius Eₐ > 0 from a slower and a faster run, residual R² changes with the model).
• CI: --cov-fail-under=81.

Review fixes on top of the original commits

• The Removal test looked for element type "imgs"; Streamlit 1.64 (what CI installs) reports "image", so it failed on Linux. It now accepts both.
• The Arrhenius test uploaded the same file for 25 °C and 60 °C (Ea = 0, nothing tested); the 60 °C file is now faster and the test asserts Ea > 0.

Verification

171 tests pass; coverage of app_ods.py 18 % → 78 %, total 45 % → 83 % (floor 81 %); ruff check / ruff format --check clean. No change to app code.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GN1ANT89MKDjHRXcnjgSjo

---

Generated by Claude Code