VecForge 2.0 experiment, benchmark and example scripts.

Install the package with: pip install "vecforge[mcp]"

Reproducing the experiments
  E1 to E7:  python run_experiments.py      (about 15 minutes on 4 cores)
  E8 to E18: export SCIFACT_DIR=/path/to/scifact     (BEIR SciFact: corpus.jsonl, queries.jsonl, qrels/)
             export MINILM_DIR=/path/to/minilm       (all-MiniLM-L6-v2 as ONNX: model.onnx, tokenizer.json)
             python run_experiments2.py              (or --only E9,E10 to rerun a subset)
  Figures:   python make_figures.py and python make_figures2.py

The SciFact dataset and the MiniLM encoder are not included. Download them separately, and check their licences.
The experiments also need onnxruntime, tokenizers and the mcp client package.

Other files
  bench_search.py, bench_quantum.py: search and reranking benchmarks.
  examples/: example applications (AI memory agent, codebase assistant, GIS data search, RAG pipeline and others).
  verify_scores.py, verify_scores_varied.py: quick checks that scores are ordered and in range.
