perf: bring musl binaries performance in par with glibc ones

After profiling, I have confirmed that the cause of the relative
slowdown of `musl` binaries with respect to `glibc` was the usage of a
slower memory allocator in the `musl` case, which critically affects the
performance of Zopfli compression, especially on CPUs with lots of
physical threads.

CHANGELOG.md

@@ -37,6 +37,18 @@ Versioning](https://semver.org/spec/v2.0.0.html).
     to their decentralized and free nature, the PackSquash project will most
     likely not pursue signing binaries with code signing certificates.
 
+#### Performance
+
+- The statically-linked PackSquash CLI Linux binaries now use the
+  [`mimalloc`](https://github.com/microsoft/mimalloc) memory allocator, instead
+  of the default `musl` C library allocator.
+  - This brings the performance of such binaries roughly in line with those
+    packaged for Debian, which dynamically link against `glibc`. Previously,
+    slowdowns of 5x or more could be expected, depending on the number of
+    threads used.
+  - Because they use statically-linked binaries, the PackSquash GitHub
+    Action and Docker container also saw performance improvements.
+
 #### Fixed
 
 - The UTF-8 BOM is no longer automatically stripped from properties files, as

packages/packsquash_cli/Cargo.toml

@@ -26,6 +26,15 @@ env_logger = { version = "0.10.0", default-features = false, features = ["color"
 winapi = { version = "0.3.9", default-features = false, features = ["consoleapi"] }
 winapi-util = "0.1.5"
 
+[target.'cfg(all(target_os = "linux", target_env = "musl"))'.dependencies]
+# The mimalloc C malloc override feature is troublesome on macOS and Windows, but
+# luckily the Linux musl C library should support this well. See:
+# https://github.com/purpleprotocol/mimalloc_rust/issues/41
+# We are interested in overriding malloc because we depend on C libraries, and doing
+# so buys us a safety net against mismatching memory managed by different allocators,
+# in addition to keep binary sizes down
+mimalloc = { version = "0.1.39", default-features = false, features = ["override"] }
+
 [target.'cfg(windows)'.build-dependencies]
 winresource = "0.1.17"
 

packages/packsquash_cli/src/bin/packsquash/main.rs

@@ -30,6 +30,15 @@ const LOG_TARGET: Target = Target::Stderr;
 /// A producer of the [`IsTerminal`] implementation that matches the [`LOG_TARGET`] constant.
 const LOG_TARGET_STREAM: fn() -> Stderr = io::stderr;
 
+/// Zopfli compression is pretty intensive on the memory allocator, and musl
+/// is known to use a simple but prone to extreme thread contention allocator.
+/// Replace its allocator with one whose performance is close to glibc so that
+/// the PackSquash CLI musl binaries have a similar performance to glibc
+/// binaries; otherwise, slowdowns of 5x or more are to be expected.
+#[cfg(all(target_os = "linux", target_env = "musl"))]
+#[global_allocator]
+static GLOBAL_ALLOCATOR: mimalloc::MiMalloc = mimalloc::MiMalloc;
+
 fn main() {
 	process::exit(run(TerminalTitleController::new()));
 }