Fix segfault when a formatted insertion produces zero characters

[afonsojanu]

Fixes #36.

cc_str_insert_wrapped_fmt_args (the function behind push_fmt and insert_fmt) first counts up the total length of everything it needs to format and insert, then, if that total exceeds the current capacity, grows the buffer, and only then does the actual memmove/insert.

The problem is when the total comes out to zero, for instance push_fmt(&s, "") on a container that hasn't allocated anything yet. cc_str_size(cntr) + total > cc_str_cap(cntr) is 0 + 0 > 0, which is false, so the growth step is skipped entirely. Execution falls straight through to the memmove and to cc_str_hdr(cntr)->size += total. For a never-allocated string, cntr still points at CC's shared placeholder for an empty string, which lives in read-only memory, so both of those touch memory that can't be written to and the process crashes.

cc_str_insert_n already has a guard for exactly this situation: it returns immediately when n == 0, before it ever looks at the container's capacity. I added the equivalent check to cc_str_insert_wrapped_fmt_args, right after the length-counting loop and before the growth step, so a zero-length formatted insertion is a no-op rather than falling through to code that assumes there's something to write.

I reproduced the crash from the issue under ASan first (SIGBUS in cc_str_insert_wrapped_fmt_args, matching what DevSolar described), confirmed the fix removes it, and added a test to unit_tests.c covering push_fmt with an empty string across all three of CC's string element types (char, char16_t, char32_t). The full unit test suite still passes clean under ASan/UBSan (including the fault-injection pass that simulates realloc failures), and tests_against_stl.cpp still compiles against the patched header.

[DevSolar]

Thanks for looking into the mechanics on how to fix this. +1 for copying a mechanic used elsewhere already.