faster path filling again

[guzba]

before:
svg decode ........................ 18.183 ms 18.360 ms ±0.167 x272

after:
svg decode ........................ 17.021 ms 17.161 ms ±0.111 x291

The win here is from 2 things:

1. Computing and storing m and b for each segment and using that for each hit instead of calculating it every time
2. Removing a float64 that snuck in through a constant (results in the file changes)