Add optional RAG cross-encoder reranking

[lmdixon23]

Summary

• add opt-in cross-encoder re-ranking over a configurable embedding candidate pool;
• preserve embedding-only retrieval as the default behavior;
• expose re-ranking through the CLI and Flask serving configuration;
• preserve the original embedding score alongside the final cross-encoder score;
• add deterministic, network-free re-ranker and pipeline tests;
• add a reproducible baseline-versus-re-ranked evaluation command;
• update sentence-transformers dimension detection for current and older supported releases.

Validation

• 19 passed in the RAG assistant test suite;
• 2 passed in the RAG-to-eval-harness bridge suite;
• real-model evaluation completed on CPU using:
  • sentence-transformers 5.6.1;
  • transformers 5.14.1;
  • torch 2.13.0.

Measured smoke result

Configuration	recall@3	MRR
Embedding only	1.000	0.900
Cross-encoder re-ranked	1.000	1.000
Observed delta	+0.000	+0.100

The re-ranker moved the approximate-nearest-neighbor case from rank 2 to rank 1.

This is a five-case smoke result over three small documents. It records the observed delta and does not establish a general quality lift. Issue #18 tracks the expanded discriminative benchmark.

Closes #10