Research: TurboQuant-enhanced vector quantization for Stoolap

Add research, use case, and planned RFCs for integrating Google
Research's TurboQuant techniques into Stoolap's vector storage:

- TurboQuant (arXiv:2504.19874): Two-stage quantization achieving
  3-bit KV cache without accuracy loss via PolarQuant + QJL

Research doc (docs/research/):
- turboquant-stoolap-enhancement.md: Technical deep-dive covering
  PolarQuant (zero-overhead polar coordinates), QJL (1-bit residual),
  random rotation (no-training PQ), and integration analysis

Use Case (docs/use-cases/):
- turboquant-vector-quantization.md: Problem statement, stakeholders,
  success metrics (≥8x compression, ≥95% recall@10), constraints

Planned RFCs (rfcs/planned/retrieval/):
- RFC-0915: TurboQuant Vector Quantization - TurboScalar (4-bit/0 const),
  ThreeBit (3-bit), TurboPQ (no-training) quantization types
- RFC-0916: TurboHNSW Quantized Index - HNSW on quantized vectors,
  dual-phase search with re-ranking, 8x memory/speed improvement

Sources:
- TurboQuant: https://arxiv.org/abs/2504.19874
- PolarQuant: https://arxiv.org/abs/2502.02617
- QJL: https://arxiv.org/abs/2406.03482
- Google Research Blog: research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/

Author: mmacedoeu