refactor: reuse boilerplate text filter

[jordansilly77-stack]

Summary

• remove the duplicated nested boilerplate-text filter in the RAG vector store path
• reuse the existing module-level _is_boilerplate_text helper for the same filtering step

Verification

• .venv/bin/python -m py_compile srcs/chatbot.py
• git diff --check
• .venv/bin/python -m pytest -q (100 passed, 2 skipped)

Closes #19

[Rekin226]

Approving — clean refactor that kills a genuine duplication.

What's right

• build_vector_store now routes through the canonical _is_boilerplate_text instead of its own private copy, so ingest-time and retrieval-time filtering finally agree.
• The two things that actually matter are preserved: the getattr(d, "page_content", "") guard and the if filtered: docs = filtered fallback (don't filter every chunk away).
• Forward reference is fine — _is_boilerplate_text is only called at runtime. py_compile + full suite green (100 passed).

One note (not blocking): this isn't byte-for-byte equivalent, so 'the same filtering step' slightly understates it:

• The old closure matched the bare substring "terms"; the canonical set uses "terms of use"/"terms and conditions", so generic 'terms' chunks now survive (fewer false positives — good).
• _is_boilerplate_text also adds the short-line-ratio heuristic the old closure lacked, so ingest filtering is stricter in that dimension.

Net effect is exactly the point — both paths now use one stricter helper. Worth a one-line test pinning that a 'terms'-only chunk now survives ingest, but that can be a follow-up. LGTM.