StringValuePtr doesn't guarantee a NULL byte at the end of the char * it
returns.  The for loop in the `parse_version_number` depends on a NULL
byte in the string in order to stop the loop.  Since `StringValuePtr`
doesn't guarantee a NULL byte in the `char *`, it's possible the for
loop could read past the end of the string, and `offset` would end up
being larger than the number of bytes that are actually in the string.

This fixes the test on Rubygems 1.8.23.2

fca9ae5 "avoid signed overflow" can still cause undefined behavior because
`compare_version_number` can return `INT_MIN` which `version_compare_cb_r`
then tries to negate:

```
../../../../ext/version_sorter/version_sorter.c:94:9: runtime error: negation of -2147483648 cannot be represented in type 'int'; cast to an unsigned type to negate this value to itself
    #0 0x7f9cde1fe1ba in version_compare_cb_r /tmp/x86_64-linux-gnu/version_sorter/2.3.1/../../../../ext/version_sorter/version_sorter.c:94:9
    #1 0x7f9cde75c231  (/lib/x86_64-linux-gnu/libc.so.6+0x39231)
    #2 0x7f9cde75c69e in qsort_r (/lib/x86_64-linux-gnu/libc.so.6+0x3969e)
    #3 0x7f9cde1fd68c in rb_version_sort_1 /tmp/x86_64-linux-gnu/version_sorter/2.3.1/../../../../ext/version_sorter/version_sorter.c:202:2
```