Micro Optimization: Elide overflow checking in Data.bytes

[jmschonfeld]

This is a tiny optimization in conversion from Data to a RawSpan that elides unnecessary overflow checks when performing arithmetic on the bounds/offsets of the bytes pointer. Given the constraints that the Range type guarantees (lowerBound <= upperBound) and the constraint that all Data indices are non-negative, we can avoid performing overflow checking on these operations since the possible inputs can never overflow. Runtime-wise the other parts of this conversion (copying the representation, switching over the cases, etc.) outweigh these overflow checks. However, I did find that removing these checks resulted in a 15% reduction in code size in the caller and as we remove the other heavy pieces these will become more prominent. As we heavily promote using Spans in API this conversion will become much more common and I think is worthy of a micro optimization like this to ensure our best performance.

[glessard]

Thanks