Fix multiple instances of undefined behaviour and crashes when using ubsan

[GasInfinity]

While compiling with zig cc I encountered crashes due to undefined behaviour (It enables ubsan by default in debug/safe builds). This PR aims to fix some code which causes ub

Some of the fixes include:

• Avoid passing NULL to src or dest in memcpy()
• Avoid incrementing/decrementing a NULL pointer
• Avoid incrementing a pointer by 0
• Avoid doing signed bit shifting by casting to unsigned first

[Aurel300]

I wonder: is this assumption ever violated? Can we use an assume intrinsic or similar?

[GasInfinity]

Well, when hashlink starts gc_roots == NULL.

frame #5: 0x00007ffff7e0abc7 libhl.so.1`hl_add_root(r=0x00007ffff7f1c1a0) at gc.c:274:3
   271          if( gc_roots_count == gc_roots_max ) {
   272                  int nroots = gc_roots_max ? (gc_roots_max << 1) : 16;
   273                  void ***roots = (void***)malloc(sizeof(void*)*nroots);
-> 274                  memcpy(roots,gc_roots,sizeof(void*)*gc_roots_count);
   275                  free(gc_roots);
   276                  gc_roots = roots;
   277                  gc_roots_max = nroots;
(lldb) v
(void *) r = 0x00007ffff7f1c1a0
(int) nroots = 16
(void ***) roots = 0x0000000001194c60
(lldb) print gc_roots
(void ***) 0x0000000000000000

So, we can add that check or initialize it before calling hl_add_root() somewhere else.

[Aurel300]

So this will now fail silently? If this check is performed, might as well emit a warning or something? Additionally, I guess this situation (bytes == NULL) is also meant to be unreachable with how the bytes wrapper works Haxe-side?

[GasInfinity]

That happened when I was testing the changes with some openfl templates. It seems that somewhere it passes NULL to hl_bytes_fill and hl_bytes_blit.

Edit: About the warning, where should it be emitted? to stderr?