compiler-rt: simplify release small memcpy

This change produces 2 fewer bytes of code but adds an unconditional
jump to the copy loop. On znver4/5 the previous implementation produced
significantly mode code (using LLVM 19).

lib/compiler_rt/memcpy.zig

@@ -34,13 +34,8 @@ comptime {
 fn memcpySmall(noalias dest: ?[*]u8, noalias src: ?[*]const u8, len: usize) callconv(.C) ?[*]u8 {
     @setRuntimeSafety(builtin.is_test);
 
-    if (len != 0) {
-        var i: usize = 0;
-        while (true) {
-            dest.?[i] = src.?[i];
-            i += 1;
-            if (i == len) break;
-        }
+    for (0..len) |i| {
+        dest.?[i] = src.?[i];
     }
 
     return dest;