feat(oprf): add revolutionary VOLE-LWR helper-less unlinkable OPRF

Implements a novel post-quantum OPRF combining:
- VOLE-based masking (prevents fingerprint attacks)
- LWR finalization (no reconciliation helpers transmitted)
- PCG pre-processing (amortized communication cost)
- NTT-friendly q=65537 (WASM performance)

Key fixes during implementation:
- LWR parameters: p=16, β=1 ensures 2nβ²=512 < q/(2p)=2048
- Password element must be UNIFORM (not small) for LWR to work
- Server subtracts v=u·Δ+noise, client just rounds (no addition)

Performance: ~82µs full protocol (vs 60µs fast, 99µs unlinkable)
Security: UC-unlinkable, helper-less, post-quantum (Ring-LWR)

All 206 tests passing.

benches/timing_verification.rs

@@ -3,12 +3,11 @@
 //! A |t-value| > 5 indicates a timing leak with high confidence.
 //! Functions should show |t-value| < 5 after sufficient samples.
 
-use dudect_bencher::{BenchRng, Class, CtRunner, ctbench_main};
+use dudect_bencher::{BenchRng, Class, CtRunner, ctbench_main, rand::Rng};
 use opaque_lattice::oprf::fast_oprf::{
     PublicParams, Q, RING_N, ReconciliationHelper, RingElement, ServerKey, client_blind,
     client_finalize, server_evaluate,
 };
-use rand::Rng;
 
 fn coin_flip(rng: &mut BenchRng) -> bool {
     rng.gen_range(0u8..2) == 0