perf: the dynamic Dory commitment computation should have a GPU imple…

…mentation (#247)

# Rationale for this change
The Blitzar project implemented a `vlen_msm` function that is optimized
for the dynamic Dory structure. This PR creates the
`dynamic_dory_commitment_helper_gpu` module when using the blitzar
feature flag. Helper functions are added that create the input needed
for Blitzar's vlen_msm function.

This work is the final piece of a PR that was split up:
#231.

# What changes are included in this PR?
- The GPU implementation of the dynamic Dory commitment computation is
implemented.
- `#[allow(dead_code)]` tags are removed from the
`dynamic_dory_structure` module.
- `pack_scalars::min_as_f` is made public - this will likely need
refactored in a follow up PR when the `compute_dynamic_T_vec_prime` is
ported to the GPU.

# Are these changes tested?
Yes

crates/proof-of-sql/src/proof_primitive/dory/dory_compute_commitments_test.rs

@@ -328,6 +328,36 @@ fn we_can_compute_an_empty_dory_commitment() {
     assert_eq!(res[0].0, GT::zero());
     let res = compute_dory_commitments(&[CommittableColumn::BigInt(&[0; 0])], 20, &setup);
     assert_eq!(res[0].0, GT::zero());
+    let res = compute_dory_commitments(
+        &[
+            CommittableColumn::BigInt(&[0; 0]),
+            CommittableColumn::BigInt(&[0; 0]),
+        ],
+        0,
+        &setup,
+    );
+    assert_eq!(res[0].0, GT::zero());
+    assert_eq!(res[1].0, GT::zero());
+    let res = compute_dory_commitments(
+        &[
+            CommittableColumn::BigInt(&[0; 0]),
+            CommittableColumn::BigInt(&[0; 0]),
+        ],
+        5,
+        &setup,
+    );
+    assert_eq!(res[0].0, GT::zero());
+    assert_eq!(res[1].0, GT::zero());
+    let res = compute_dory_commitments(
+        &[
+            CommittableColumn::BigInt(&[0; 0]),
+            CommittableColumn::BigInt(&[0; 0]),
+        ],
+        20,
+        &setup,
+    );
+    assert_eq!(res[0].0, GT::zero());
+    assert_eq!(res[1].0, GT::zero());
 }
 
 #[test]