A weird order8(?) CM coder

[Shelwien]

http://ctxmodel.net/files/mix_test/mix_test_vB.rar

This is not especially good in any sense, but is kinda unique still.
Because imho its a first CM implementation with only direct access
to all (masked) context statistics and with CM-like compression ratio.
That is, there're no hashes or trees - only static tables where
each element corresponds to a specific context.
Context masks are pretty funny - check the .idx files in archive.

[ mix_test 09h2x4a CM coder (87M stats memory) ]
book1.rc: 222010
enwik6.rc: 252742
wcc386.rc: 269768

[ mix_test 09h2x4b CM coder (310M stats memory) ]
book1.rc: 218899
enwik6.rc: 248272
wcc386.rc: 267258

[Shelwien]

http://ctxmodel.net/files/mix_test/mix_test_vC.rar
+ 09h2x4a coder fully tuned (244M)
+ alphabet reordering test
+ delayed counter test

book1  enwik6 wcc386 bookstar enwik8   enwik9    
209675 249679 290768  9903255 23561835 208226850 // 09h2x4a: bitfield context mixing
206927 242855 304590 10815006 22757931 200820030 // 09i0: alphabet reordering
208994 245859 288623  9760519 23338466 206267385 // 09i2a: delayed counters
206269 239101 302402 10908714 22552419 199084429 // 09i2b CM coder : AR + DC
208208 250792 278422  9098171 20608793 167852106 // BWTmix v1

Somehow this doesn't scale very well, it seems.
For example, 09h2x4a enwik8-9 results seem similar to that of m1, which only
has two counters.