Better handling for number dictionary

[nsmarkop]

Unit testing the numbers dictionary has revealed some issues even though the base functionality works as expected.

First, reverse_lookup sort-of worked but does not handle cases where the numbers are split with - or reversed with * properly. This may be a lost cause as at the moment it seems like it'd take significant effort to resolve.

Secondly, lookup is quite permissive of input. If you give it 1258 instead of 125-8, 1290 instead of 12-90, etc. it works fine even though that violates how the steno order is supposed to be written.

Thirdly, You can't unit test reverse_lookup code that has a return value (so, a "passing" case) because when it calls normalize_stroke it errors out due to a raised AssertionError because a "system" has not been initialized in the plover package which it's using during normalize_steno internally for things like NUMBER_KEY lookups and such.