fix(stapp): fall back to st.rerun() when fragment scope rerun is unavailable

[Kailigithub]

Summary

Fixes #564

The Streamlit sidebar in frontends/stapp.py crashes on startup when the persisted LLM selectbox value (st.session_state['sidebar_llm_select']) disagrees with the agent's current llm_no:

StreamlitAPIException: scope="fragment" can only be specified from
@st.fragment-decorated functions during fragment reruns.

Root cause

render_sidebar() is decorated with @st.fragment. On the initial app run, the function executes as part of the main script — not as a fragment rerun. If selected_idx != current_idx triggers in that first execution, calling st.rerun(scope="fragment") is illegal and Streamlit aborts.

The other st.rerun(scope="app") calls in the same fragment are unaffected — scope="app" is legal from both fragment and non-fragment contexts; the restriction applies only to scope="fragment".

Fix

Wrap the call so that when fragment-scoped rerun is unavailable (initial run / outside a fragment rerun), we fall back to a full st.rerun(). The user-visible behavior is unchanged: the LLM is switched and the UI refreshes either way.

if selected_idx != current_idx:
    agent.next_llm(selected_idx)
    try:
        st.rerun(scope="fragment")
    except st.StreamlitAPIException:
        st.rerun()

frontends/stapp2.py does not use scope="fragment" and is unaffected. No other frontend touches this pattern.

Verification

• python3 -m py_compile frontends/stapp.py — passes
• ruff check frontends/stapp.py — no new findings introduced by this change (60 pre-existing project-style warnings remain, untouched)
• Manual review: try/except is the documented Streamlit pattern for this scenario; behavior on subsequent fragment reruns (where scope="fragment" is legal and faster) is preserved