Add smoke test and benchmarks

smoke links crt0.o + libc.a with -nostdlib -static and proves the whole chain runs without the dynamic linker; bench_strlen and bench_syscall measure the string core and raw syscall round-trip cost via make bench.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <[email protected]>

benchmarks/bench_syscall.c

@@ -0,0 +1,29 @@
+/*
+ * bench_syscall.c — raw syscall round-trip cost.
+ *
+ * This is the number that justifies the whole project: for a kernel-first
+ * libc (project guideline #6), every wrapper is one `syscall` instruction
+ * away from the kernel. getpid() is the cheapest syscall there is, so
+ * this measures the floor. 1M iterations; results in ns per call.
+ */
+
+#include <unistd.h>
+
+#include "bench.h"
+
+#define ITERS 1000000UL
+
+int main(void)
+{
+    volatile pid_t sink = 0;
+    long long t0, t1;
+
+    t0 = bench_now_ns();
+    for (unsigned long i = 0; i < ITERS; i++)
+        sink += getpid();
+    t1 = bench_now_ns();
+
+    (void)sink;
+    bench_print_u64((unsigned long long)((t1 - t0) / ITERS));
+    return 0;
+}