[CIR][CodeGen] fixes access to globals with bitfields

[gitoleg]

This PR adds a bitcast when we rewrite globals type. Previously we just set a new type and it worked.
But recently I started to test ClangIR with CSmith in order to find some run time bugs and faced with the next problem.

typedef struct {
    int x : 15;   
    uint8_t y;
} S;

S g = { -12, 254};

int main() {    
    printf("%d\n", g.y);
    return 0;
}

The output for this program is ... 127 but not 254!
The reason is that first global var is created with the type of struct S, then get_member operation is generated with index 1
and then after, the type of the global is rewritten - I assume because of the anon struct created on the right side in the initialization.
But the get_member operation still wants to access to the field at index 1 and get a wrong byte.
If we change the y type to int we will fail on the verification stage. But in the example above it's a run time error!

This is why I suggest to add a bitcast once we have to rewrite the global type.

[bcardosolopes]

Nice catch!