Protect against different vector widths (#1909)

GCC aligns alloca to the widest enabled vector type (16 bytes, 32 with avx,
64 with avx-512). With AVX the block address is `lea 0x1f(%rsp)` rounded down
to the nearest 32-divisible, but half the time the block starts 16 bytes higher
so the function reads the stack arguments 2 slots away from where it should.
Without AVX the address is `lea 0xf(%rsp)` which is the stack pointer.

src/core/ffi.c

@@ -1225,7 +1225,15 @@ static Janet janet_ffi_sysv64(JanetFFISignature *signature, void *function_point
         ret_mem = alloca(type_size(signature->ret.type));
         regs[0] = (uint64_t) ret_mem;
     }
+/* GCC aligns alloca to the widest allowed vector type (32 bytes for AVX) which can pad
+ * between the stack pointer and this block, but stack args start at the stack pointer,
+ * so we must ask for the calling convention's 16 byte alignment */
+/* Both GCC and CLANG define GNUC. */
+#if defined(__GNUC__)
+    uint64_t *stack = __builtin_alloca_with_align(sizeof(uint64_t) * signature->stack_count, 128);
+#else
     uint64_t *stack = alloca(sizeof(uint64_t) * signature->stack_count);
+#endif
     for (uint32_t i = 0; i < signature->arg_count; i++) {
         uint64_t *to;
         int32_t n = i + 2;