Introduce constant-time primitive for absolute value computation

`poly_chknorm` uses the pattern

```c
    t = a->coeffs[i] >> 31;
    t = a->coeffs[i] - (t & 2 * a->coeffs[i]);
```

to compute the absolute value of an `int32_t`. This should be wrapped in a constant-time function guarded by barriers making compilation into a branch unlikely.