Pull starting slice from redis instead of extrapolating

[sbwdlihao]

fix #146

[coveralls]

Coverage increased (+0.01%) when pulling 1c51806 on sbwdlihao:patch-3 into 50ceb67 on simplecrypto:master.

[icook]

I'm unsure about the correctness of this. In scenarios involving merge-mined blocks this will definitely be wrong at certain times. Effectively, something can occur where "slice 0" has most of the shares, and was rotated by a merge mined block solve. "slice 1" has a small number of shares, and was rotated due to a main network block solve.

The simplest fix is to require the "start_index" value if there are other blocks solved on the same chain already. This has it's own problems tho if the redis database gets wiped.

I'll keep thinking on this. For the large majority of cases tho this should work fine. Thanks for the work!