perf(backends/arkworks): batch normalization + rayon-parallel DoryRoutines

CHANGELOG.md

@@ -5,6 +5,25 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+### Added
+
+- Safe conversions between the arkworks wrappers and their inner types:
+  `into_inner()` plus `From` impls in both directions for
+  `ArkFr`/`ark_bn254::Fr`, `ArkG1`/`G1Projective`, and `ArkG2`/`G2Projective`.
+
+### Changed
+
+- Faster arkworks `DoryRoutines`: when MSM bases are not already normalized
+  (e.g. folded vectors mid reduce-round), `msm` converts them to affine with a
+  single batched inversion (`normalize_batch`) instead of one field inversion
+  per point; already-normalized bases (setup generators) keep the
+  inversion-free per-element path. `fixed_base_vector_scalar_mul`,
+  `fixed_scalar_mul_bases_then_add`, `fixed_scalar_mul_vs_then_add`, and
+  `fold_field_vectors` are parallelized under the `parallel` feature. Group
+  results are unchanged, so commitments and proofs are byte-identical.
+
 ## [0.4.0] - 2026-07-03
 
 ### Security

src/backends/arkworks/ark_field.rs

@@ -12,6 +12,28 @@ use ark_std::ops::{Add, Mul, Neg, Sub};
 #[repr(transparent)]
 pub struct ArkFr(pub Fr);
 
+impl ArkFr {
+    /// Returns the wrapped `ark_bn254::Fr`.
+    #[inline]
+    pub fn into_inner(self) -> Fr {
+        self.0
+    }
+}
+
+impl From<Fr> for ArkFr {
+    #[inline]
+    fn from(value: Fr) -> Self {
+        ArkFr(value)
+    }
+}
+
+impl From<ArkFr> for Fr {
+    #[inline]
+    fn from(value: ArkFr) -> Self {
+        value.0
+    }
+}
+
 impl Field for ArkFr {
     fn zero() -> Self {
         ArkFr(Fr::from(0u64))