viewport.rs: in tests, use numbers that can be represented exactly as f64
In clip_to_viewport(), we were expecting a result of 0.2, which can't be represented exactly. So, the resulting matrix with 0.2 components in it wasn't being compared successfully on i386. (why was it fine on x86_64?) We'll need some bigger guns for asserting on float "equality", it seems.
Showing with 6 additions and 6 deletions