perf(bench): add optional mimalloc allocator for sp1-perf (-7.2% fib)

Add mimalloc as an optional feature for sp1-perf. When enabled,
replaces glibc malloc with mimalloc's per-thread arenas, reducing
lock contention in allocation-heavy paths.

Benchmark results (on top of physical-cores change):
  fib:    28,551ms → 26,504ms  (-7.2%)
  keccak: 35,648ms → 34,845ms  (-2.3%)
  big:    38,228ms → 37,899ms  (-0.9%)

Usage: cargo build --release -p sp1-perf --features mimalloc

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: tamirhemo

Cargo.lock

@@ -33,6 +33,7 @@ tracing = { workspace = true }
 serde = { workspace = true }
 serde_json = { workspace = true }
 test-artifacts = { path = "../test-artifacts" }
+mimalloc = { version = "0.1", optional = true }
 
 [[bin]]
 name = "sp1-perf"

crates/perf/Cargo.toml

@@ -1,3 +1,7 @@
+#[cfg(feature = "mimalloc")]
+#[global_allocator]
+static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;
+
 use std::{
     future::Future,
     time::{Duration, Instant},