Fix height/width computation

[jonasteuwen]

This PR fixes the height/width computation. See #20. Also matches python implementation.

[jonasteuwen]

This fixes level 0, but not the rest (openphi vs libisyntax):

1. (202758, 95238) - (202758, 95238)
2. (101377, 47617) - (101379, 47619) (+2)
3. (50686, 23806) - (50689, 23809) (+3)
4. (25341, 11901) - (25344, 11904) (+3)
5. (12669, 5949) - (12672, 5952) (+3)
6. (6333, 2973) - (6336, 2976) (+3)
7. (3165, 1485) - (3168, 1488) (+3)
8. (1581, 741) - (1584, 744) (+3)
9. (789, 369) - (792, 372) (+3)

Note level_padding is 1533 in this case.
This is very weird though, you'd expect a / 2.

[Falcury]

Can I merge this, or should we wait until it's fixed for all the levels beside level 0?

[jonasteuwen]

@Falcury I would fix it in level 0, the other levels are no proper power of 2, and I can’t figure out any formula that generates these values.

These 2-3 pixels are also not an issue because of the minimal padding, so I don’t think it will ever be out of bounds. So I would consider the current one ‘correct’