use a more optimal loop implementation (#131)

In platform-tools v1.53 inlining decisions of Iterator::next prevent
LLVM from optimizing this loop into an unrolled index-based version of
it.

So where in 1.51 we'd see

```asm
INLINE_MEMMOVE($from, $to, 32 bytes)
jeq $length, 1, exit
INLINE_MEMMOVE($from, $to, 32 bytes)
jeq $length, 2, exit
...
```

In 1.53 we'd instead get a much more CU-heavy pointer-comparison based
loop instead.

```asm
lsh64 $index, 3
ldxdw r4, [r1 + 0]
add64 r1, $index
INLINE_MEMMOVE($from, $to, 32)
ldxdw r6, [r10 - 48]
jeq r1, r2, exit
; ...
```

Since we want an index-based loop here for optimal results, might as
well write it that way directly instead of relying on optimizations to
trigger.

Co-authored-by: Lucas Ste <38472950+LucasSte@users.noreply.github.com>

Author: nagisa

pinocchio/program/src/processor/shared/initialize_multisig.rs

@@ -66,8 +66,9 @@ pub fn process_initialize_multisig(
         return Err(TokenError::InvalidNumberOfRequiredSigners.into());
     }
 
-    for (i, signer_info) in remaining.iter().enumerate() {
-        multisig.signers[i] = *signer_info.key();
+    #[expect(clippy::needless_range_loop)] // CU use is better with index-based loop
+    for i in 0..remaining.len() {
+        multisig.signers[i] = *remaining[i].key();
     }
 
     multisig.set_initialized(true);