Add comprehensive ANN benchmarking suite

Extend benchmarks-ann/ with results database (SQLite with per-query detail
and continuous writes), dataset subfolder organization, --subset-size and
--warmup options. Supports systematic comparison across flat, rescore, IVF,
and DiskANN index types.

benchmarks-ann/datasets/nyt/Makefile

@@ -0,0 +1,30 @@
+MODEL ?= minishlab/potion-base-8M
+K ?= 100
+BATCH_SIZE ?= 512
+DATA_DIR ?= data
+
+all: base.db contents.db
+
+# Download NYT headlines CSVs from Kaggle (requires `kaggle` CLI + API token)
+$(DATA_DIR):
+	kaggle datasets download -d johnbandy/new-york-times-headlines -p $(DATA_DIR) --unzip
+
+contents.db: $(DATA_DIR)
+	uv run build-contents.py --data-dir $(DATA_DIR) -o $@
+
+base.db: contents.db queries.txt
+	uv run build-base.py \
+		--contents-db contents.db \
+		--model $(MODEL) \
+		--queries-file queries.txt \
+		--batch-size $(BATCH_SIZE) \
+		--k $(K) \
+		-o $@
+
+clean:
+	rm -f base.db contents.db
+
+clean-all: clean
+	rm -rf $(DATA_DIR)
+
+.PHONY: all clean clean-all