Previously it returned `undef`.
So for example, if there was a syntax error, the following code would cause `SIGSEGV`.

```c
const char *code = "syntax ? error ?";
mrb_p(mrb, mrb_load_string(mrb, code));
```

This affects the behavior of the `MRB_API` function below:
- `mrb_load_file()`
- `mrb_load_file_cxt()`
- `mrb_load_detect_file_cxt()`
- `mrb_load_string()`
- `mrb_load_nstring()`
- `mrb_load_string_cxt()`
- `mrb_load_nstring_cxt()`

I mentioned earlier why I want to change from `undef` to `nil`, but here's why I decided to change it:
- Returns `nil` if an error occurred when returning from `mrb_top_run()`.
- The specific error can be found by confirming that `mrb->exc` is non-`NULL`.
- Even if `mrbc_context::no_exec` is true, it is sufficient to check if it is either `proc` or `nil`.

However, this change can cause compatibility issues if it has already used `mrb_undef_p()` to distinguish between syntax errors.
This is the reason for making changes to the `mrdb.c` file (although the behavior does not change with or without changes).