Fix resolve_logfield_string()

When ATS receives a malicious request which URL is too long to hold by
internal_msg_buffer, the internal_msg_buffer_size might be set to 0.

As a result, the appended memory which allocated by ats_malloc() would
be mistaken for the memory from ink_freelist, and would be free to
ink_freelist finally.

As this memory is larger than the one in ink_freelist, and all memory in
the origin ink_freelist would not be reclaimed, so it wouldn't cause
segment-fault, that is why we didn't notice it in the past.

But after we use reclaimabe-freelist, this bug would cause segment-fault
when use it to get inner meta-data or free it back to OS by unmmap().

===
Now, we found the root cause which would lead to internal_msg_buffer_size to 0
while internal_msg_buffer is NOT NULL.

That is resolve_logfiled_string() function. Let's fix it.