fix: reject batch_size < 1 and handle empty-query benchmarks

[HarshRajSinghania]

Summary

Validate ingest batch size and return empty-mode results for a zero-query benchmark instead of crashing.

Fixes #22 and #23.

Motivation

ingest_documents(..., batch_size=0) forwarded a zero step into range(), which raised ValueError: range() arg 3 must not be zero.
run_benchmark divided by len(r5) even when the dataset had no queries, which raised ZeroDivisionError.

Both were labeled good first issue / help wanted, were unassigned, and had no open PR.

Implementation

• ingest_documents raises ValueError("batch_size must be >= 1") before batching.
• batched() uses the same guard so other callers cannot hit the raw range() error.
• run_benchmark treats an empty query list as recall 0.0 / qps 0.0 and still returns a ModeResult per requested mode (n_queries=0).

Testing

Installed pytest, qdrant-client, and prometheus-client in an isolated environment (full pip install -e .[dev] failed on a hatchling index 502).

PYTHONPATH=src python3 -m pytest tests/test_ingest.py tests/test_benchmark.py::test_run_benchmark_empty_queries_returns_empty_modes tests/test_benchmark.py::test_percentile_and_recall_helpers -q

Result: 6 passed.

Not run: full suite (test_search.py / test_benchmark_produces_numbers) because those need the project's FastEmbed optional stack and a longer install that was blocked here.

Notes

No public API shape change besides the new validation error message requested in #22.