Handle unique edge case offered by '-option numpad:mac'

[jordansissel]

New test added. Typing tests pass.

In this case, the KEYPAD type deletes(?) its Level 1 by setting
map[None] = 2 and level_name[2] = "Number" which appears to make
level_name[1] null.

The weird edge case here is that "no modifiers" is mapped to a
non-level-1 shift level. And further, that the level entry seems to just be deleted in the keymap, no name, no atom, etc.

I confirmed this causes keypad keys to have their
key_type->level_names[0] == 0.

So now, levels with level_names[level] == 0 are skipped.

I also tried rewriting the keymap scanning code to be a little cleaner?
I removed the 'li == 0' shortcut because in this specific case, li==0 ("shift level
1") doesn't actually have any mappings and causes xdo to error when
trying to look up the level name! It's null! It has no name!

Fixes #507

[jordansissel]

Since this PR changes how the keymap is processed, it's entirely possible that this fixes #507 while breaking something else.

[dorsiflexion]

Can confirm the fix. I tried a few commands and didn't find any breakage so far.