Dividing BitLen by math.Log2(10) is what math/big does underneath

Not including the Int64/Uint64 check makes this slightly slower than old method

Included 2 benchmarks, for 10 digit numbers & 100 digit numbers:

-- before
> go test -bench=NumDigit -run=NumDigit
goos: linux
goarch: amd64
pkg: github.com/shopspring/decimal
cpu: AMD Ryzen 7 7840U w/ Radeon  780M Graphics
BenchmarkDecimal_NumDigits10-16     	18317293	        63.87 ns/op
BenchmarkDecimal_NumDigits100-16    	 3645015	       329.6 ns/op

-- after
...
BenchmarkDecimal_NumDigits10-16     	143781325	         8.488 ns/op
BenchmarkDecimal_NumDigits100-16    	 5931247	       207.4 ns/op

9007199254740992 converts to float64