implement Log10 and CmpUint64 (alternate version)

[holiman]

Alternative to #134

This PR implements Log10 in what is probably close to the optimal way. Both PRs handle the majority of cases by one lookup based on the bitlength.

For the remaining cases, #134 does 0, or, 1, or 2 divisions, and then a few uint64-compares and then it's done.

In this PR, for the remaining cases, we do an additional lookup, and then a comparison, and then we're done.

The drawback of this PR is that it has more precalculated 'persistent memory' hog.

pows = [60]Int         // 1920B 
pows64 = [20]uint64 // 20*8 = 160B

So around 2 K. Not much, but still

[holiman]

From CI benchmark run

#134 :

BenchmarkLog10/Log10/uint256-36                     	  168109	      6829 ns/op	       0 B/op	       0 allocs/op
BenchmarkLog10/Log10/big-36                         	   36321	     34610 ns/op	   23424 B/op	     510 allocs/op

This PR

BenchmarkLog10/Log10/uint256-36                     	  705776	      1597 ns/op	       0 B/op	       0 allocs/op
BenchmarkLog10/Log10/big-36                         	   27064	     44892 ns/op	   23424 B/op	     510 allocs/op

So this is on the order of 3-4x faster.

[codecov]

Codecov Report

Merging #136 (fc79e38) into master (1c9ad56) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##            master      #136   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            5         5           
  Lines         1628      1646   +18     
=========================================
+ Hits          1628      1646   +18