refactor md5.body function to reduce instruction count

- replaced manual unrolling with loop structures and constant arrays
- instruction count reduced from 12445 to 4016
- maybe about 1 to 2% performance loss on some benchs but take this
number with a grain of salt.

benchmarks/stdlib/hash/md5.c3

@@ -0,0 +1,41 @@
+module md5_bench;
+
+fn void initialize_bench() @init
+{
+	set_benchmark_warmup_iterations(3);
+	set_benchmark_max_iterations(128);
+
+    input = mem::alloc_array(char, BUFSZ);
+    input[:BUFSZ] = (char[]){ [0..BUFSZ-1] = 0xA5 }[..];
+    input_slice = input[:BUFSZ];
+}
+
+fn void teardown_bench() @finalizer
+{
+    mem::free(input);
+    input = null;
+}
+
+char* input;
+char[] input_slice;
+const usz BUFSZ = 1024 * 1024;
+
+module md5_bench @benchmark;
+
+import std::hash;
+
+fn void md5_hash()
+{
+    runtime::@start_benchmark();
+    char[*] myset = md5::hash(input_slice);
+    runtime::@end_benchmark();
+    mem::zero_volatile(myset[..]);
+}
+
+fn void compared_with_sha256()
+{
+    runtime::@start_benchmark();
+    char[*] myset = sha256::hash(input_slice);
+    runtime::@end_benchmark();
+    mem::zero_volatile(myset[..]);
+}