feat: keyword search over topics in the prospective demo

[picaultj]

Summary

Implements "search among topics by keywords" — the third idea in upstream issue rte-france/BERTrend#59 — for the prospective demo's signal-analysis view.

What it does

Adds a keyword search box above the weak/strong/noise topic tables. Typing filters all three tables to the topics whose title, LLM description, or representation match the query.

• Case-insensitive substring matching.
• Whitespace-separated terms are combined with AND (a topic must contain every term).
• Blank query → tables shown unchanged.
• A caption echoes the active filter when a query is present.

Changes

• topic_search.py (new): pure, dependency-light helper filter_topics_by_keywords(df, query, search_columns=None). No Streamlit dependency, so it is unit-testable in isolation.
• dashboard_signals.py: adds the st.text_input search box and routes each category dataframe through the helper before display.
• i18n_translations.py: FR/EN strings (search_topics, search_topics_placeholder, search_active).
• tests/apps/test_topic_search.py (new): unit tests for the helper.

Scope notes

• Kept to the signal-analysis topic tables (the primary "topics" surface). The source-explorer selectbox in the right column is left as-is; extending the filter there (and to the comparative dashboard) is a natural follow-up.
• This covers only the "keyword search" bullet of Some ideas rte-france/BERTrend#59; the other ideas (topic feedback, geographical filtering, better weak-signal characterisation) are separate, larger pieces.

Tests

tests/apps/test_topic_search.py: 9 passed (blank/None/empty handling, case-insensitive matches on title/description/representation, AND semantics, no-match, missing columns, custom search_columns).

---

Generated by Claude Code