conversion: add PrettyDec, optimize Dec, add benchmarks and fuzzing

[holiman]

This PR implements the Dec method 'natively', and adds the API-method PrettyDec.

// Dec returns the decimal representation of z.
func (z *Int) Dec() string

// PrettyDec returns the decimal representation of z, with thousands-separators.
func (z *Int) PrettyDec(separator byte) string 

It also adds benchmarks and a new fuzz-test for the changed methods.

BenchmarkDecimal/ToDecimal/uint256-8         	   12324	    102673 ns/op	   11344 B/op	     248 allocs/op
BenchmarkDecimal/ToPrettyDecimal/uint256-8   	    9595	    110235 ns/op	   14720 B/op	     251 allocs/op
BenchmarkDecimal/ToDecimal/big-8             	   10000	    133930 ns/op	   31920 B/op	     594 allocs/op

[codecov]

Codecov Report

Merging #130 (e637640) into master (87b9142) will not change coverage.
The diff coverage is 100.00%.

❗ Current head e637640 differs from pull request most recent head 46c05a7. Consider uploading reports for the commit 46c05a7 to get more accurate results

@@            Coverage Diff            @@
##            master      #130   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            5         5           
  Lines         1543      1597   +54     
=========================================
+ Hits          1543      1597   +54     

[holiman]

The 'native' implementation is better on allocs, but way slower

BenchmarkDecimal/ToDecimal/uint256-8         	    1590	    745711 ns/op	   11344 B/op	     248 allocs/op
BenchmarkDecimal/ToPrettyDecimal/uint256-8   	    2836	    749668 ns/op	   14720 B/op	     251 allocs/op
BenchmarkDecimal/ToDecimal/big-8             	    9585	    144120 ns/op	   31920 B/op	     594 allocs/op

[holiman]

Now it looks beter:

goos: linux
goarch: amd64
pkg: github.com/holiman/uint256
cpu: 12th Gen Intel(R) Core(TM) i7-1270P
BenchmarkDecimal/ToDecimal/uint256-8         	   10176	    126366 ns/op	   11344 B/op	     248 allocs/op
BenchmarkDecimal/ToPrettyDecimal/uint256-8   	    6853	    155594 ns/op	   40784 B/op	     754 allocs/op
BenchmarkDecimal/ToDecimal/big-8             	   10000	    147628 ns/op	   31920 B/op	     594 allocs/op
PASS
ok  	github.com/holiman/uint256	5.801s

[holiman]

BenchmarkDecimal/ToDecimal/uint256-8         	   12324	    102673 ns/op	   11344 B/op	     248 allocs/op
BenchmarkDecimal/ToPrettyDecimal/uint256-8   	    9595	    110235 ns/op	   14720 B/op	     251 allocs/op
BenchmarkDecimal/ToDecimal/big-8             	   10000	    133930 ns/op	   31920 B/op	     594 allocs/op

Now both of them are faster (and less alloc:y) than the big.Int decimal-print.

[holiman]

• fuzzing for them