Avoid redundant bounds checks

[kornelski]

        self.slice[self.pos] = first;
        self.slice[self.pos + 1] = second;
        self.slice[self.pos + 2] = third;
        self.slice[self.pos + 3] = fourth;

Code like this performs bounds check four times, and generates panicking code for four unique locations, which is relatively bloated for something as simple as writing four bytes.

OTOH assigning to a slice that is known to be 4 bytes long doesn't need extra bounds checks. split_at_mut(4) can create such slice, and advance the position in one go.

I've also eliminated separate tracking of start position via pos field by advancing the slice itself. To get the amount of bytes written I use a pointer comparison. It doesn't require unsafe, because the pointer is only used as an integer, and not as a pointer, and the address difference is guaranteed to be valid for positions within a single slice.

https://rust.godbolt.org/z/o95vE7rWT

[hsivonen]

Thank you!