fix: skip index build when a source produces no text chunks

[pcbeingused333]

Problem

Retriever.index() appends chunks and then always builds the backend:

for doc in docs:
    for chunk in chunk_text(doc, mode=self.chunk_mode):
        self.chunks.append(chunk)

if self.use_faiss:
    self._build_faiss()

A source that yields no text — an empty file, a scanned/image-only PDF (_read_pdf returns None, load_documents filters it out), or a blank string — leaves self.chunks empty. _build_faiss() then calls _embed([]) and reads vectors.shape[1]:

>>> RAG(use_faiss=True).add_source("scanned.pdf")
IndexError: tuple index out of range          # openai-compatible embedder
# or
ValueError: need at least one array to concatenate   # local embedder

Querying before a successful index also raises from _get_faiss() rather than returning nothing.

Fix

• index() logs a warning and returns early when self.chunks is empty after chunking.
• _query_faiss() returns [] when self._faiss_index is None.

No change when there is real text to index. index([]) on the BM25 path already worked and still does.

Tests

Added TestIndexWithoutChunks: FAISS index([]) and index([" \n\n "]) don't build, BM25 index([]) doesn't raise, and query() before any index returns []. The FAISS cases fail on main. nox -t fenn is green (922 passed, 7 skipped); ruff check / ruff format clean.

[pcbeingused333]

The pre-commit check is red on the ty hook, but that's pre-existing on main — commit 0d1862e (current main HEAD) fails the same job, on fenn/experimental/vision/normalize.py and fenn/logging.py, neither of which this PR touches. ruff check, ruff format and typos pass, and all four test (3.11–3.14) jobs are green.