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/cohere10m/Makefile

@@ -0,0 +1,27 @@
+BASE_URL = https://assets.zilliz.com/benchmark/cohere_large_10m
+
+TRAIN_PARQUETS = $(shell printf 'train-%02d-of-10.parquet ' 0 1 2 3 4 5 6 7 8 9)
+OTHER_PARQUETS = test.parquet neighbors.parquet
+PARQUETS = $(TRAIN_PARQUETS) $(OTHER_PARQUETS)
+
+.PHONY: all download clean
+
+all: base.db
+
+# Use: make -j12 download
+download: $(PARQUETS)
+
+train-%-of-10.parquet:
+	curl -L -o $@ $(BASE_URL)/$@
+
+test.parquet:
+	curl -L -o $@ $(BASE_URL)/test.parquet
+
+neighbors.parquet:
+	curl -L -o $@ $(BASE_URL)/neighbors.parquet
+
+base.db: $(PARQUETS) build_base_db.py
+	uv run --with pandas --with pyarrow python build_base_db.py
+
+clean:
+	rm -f base.db