Make 7x7 non_zero count and zig49 a single array

[mcroomp]

• Incrementally advance through array as we update
• Align counts at 256 byte boundary (types are already close to that)
• Split out zero prior_bit_len since this is many times sequentially accessed, it also makes the counts just under 256 bytes (as opposed to 0x106 bytes)

[Melirius]

Hmm, on my machine it does not improve performance.

[mcroomp]

Not sure why it's slower... oh well. Maybe I missed something in the way code generation works.