Previously they were only supported at the top-level.
Incorrect parsing lead these to be treated as a constant of the lower
bound.

Fixes #172

better coverage between top-level vs signed_ints member for various
ranges that map to primitives

fix for when >isize::MAX limits are used as we used `isize` when `i128`
would be better for covering all possibilities (`i64`, `u64`, `f64`)

float ranges now panic when they have a decimal part as we were silently
ignoring this. Issue created: #178