vector: fix support for odd-sized structures

[rr-]

During implementing a fix for LostArtefacts/TRX#1463, I've discovered a problem with our design of the Vector module. Since the vector items are now implemented as void* pointers and we encourage a direct interaction with vector->items by the user, it turns out that the only effective item_size is that of a (void*) pointer.

I've changed the vector internals to use a linear byte buffer underneath and to require always passing values around through a pointer. This makes it a bit more cumbersome to use for primitive types:

VECTOR *const vec = Vector_Create(sizeof(int32_t));
-Vector_Add(vec, (void*)5);
+Vector_Add(vec, &(int32_t){5});

But the new version allows us to use the vector with structures:

typedef struct {
    int16_t a, b;
} COMPLEX;

int main(int argc, char **argv)
{
    VECTOR *v = Vector_Create(sizeof(COMPLEX));
    Vector_Add(v, &(COMPLEX){.a = 2, .b = 4});
    Vector_Add(v, &(COMPLEX){.a = 1, .b = 3});

    for (int32_t i = 0; i < v->count * v->item_size; i++) {
        printf("%02x ", (*(char**)v->priv)[i]);
    }
    printf("\n");

    COMPLEX a = *(COMPLEX*)Vector_Get(v, 0);
    COMPLEX b = *(COMPLEX*)Vector_Get(v, 1);
    printf("%d %d\n", a.a, a.b);
    printf("%d %d\n", b.a, b.b);
}

Output:

02 00 04 00 01 00 03 00
2 4
1 3

Fortunately we don't use it anywhere so there's no need to test for regressions.
I'm not sure about whether I've implemented every function correctly, though, and would appreciate additional verification – I did test them, but I'm just not confident I've covered every corner case.

[walkawayy]

The code looks good to me.

[rr-]

Merged manually in c89eb53.