Start replacing Base's _reinterpret with our implementation.
Perf improvement on Base.padding (and Base.packedsize):
while-loop implementation of base.padding. Makes `padding()` about 4x faster:
```
julia> @btime Base.padding($(type1(30)));
34.125 μs (1175 allocations: 119.58 KiB)
julia> @btime Base.padding($(type1(30)));
8.070 μs (170 allocations: 10.72 KiB)
```