Remove unnecessary decimal rescaling - memory usage optimization

[mwoss]

I've tried to fix the performance issue mentioned in #81. Partially succeed as I couldn't simply return an already existing decimal object in rescale method due to some problems with DivMod pointer receiver method that change decimal's value attribute. As a workaround, I've created RescalePair helper function that rescales only one or none decimals from pair to avoid unnecessary memory allocations.

Results of optimization shown below.

I've used benchmarks implemented by @ngalaiko in PR #102. Thanks for the contributions to the library, I really appreciate it.

Before:

Benchmark_decimal_Decimal_Add_different_precision-8      3753075               317 ns/op             605 B/op         12 allocs/op
Benchmark_decimal_Decimal_Sub_different_precision-8      3879166               308 ns/op             557 B/op         12 allocs/op
Benchmark_math_big_Float_Sub_different_precision-8      372282800                3.24 ns/op            2 B/op          0 allocs/op
Benchmark_math_big_Float_Add_different_precision-8      37566445                31.4 ns/op            26 B/op          0 allocs/op
Benchmark_decimal_Decimal_Add_same_precision-8           6109453               201 ns/op             349 B/op          8 allocs/op
Benchmark_decimal_Decimal_Sub_same_precision-8           6106828               197 ns/op             349 B/op          8 allocs/op
Benchmark_math_big_Float_Add_same_precision-8           18792164                65.3 ns/op           105 B/op          1 allocs/op
Benchmark_math_big_Float_Sub_same_precision-8           18800349                65.5 ns/op           103 B/op          1 allocs/op

After:

Benchmark_decimal_Decimal_Add_different_precision-8      4774314               253 ns/op             493 B/op          9 allocs/op
Benchmark_decimal_Decimal_Sub_different_precision-8      4920571               243 ns/op             446 B/op          9 allocs/op
Benchmark_math_big_Float_Sub_different_precision-8      458631969                2.62 ns/op            2 B/op          0 allocs/op
Benchmark_math_big_Float_Add_different_precision-8      37542704                31.3 ns/op            26 B/op          0 allocs/op
Benchmark_decimal_Decimal_Add_same_precision-8          20565834                59.5 ns/op           130 B/op          2 allocs/op
Benchmark_decimal_Decimal_Sub_same_precision-8          20362555                61.1 ns/op           131 B/op          2 allocs/op
Benchmark_math_big_Float_Add_same_precision-8           18800202                65.7 ns/op           103 B/op          1 allocs/op
Benchmark_math_big_Float_Sub_same_precision-8           18510948                65.4 ns/op           103 B/op          1 allocs/op

[mwoss]

Tests failing on Go1.2.2, I'll fix it tomorrow.

[mwoss]

Removed benchmarks using big.NewFromFloat (my bad adding them). I'll leave comparison table in the main commnet.

[njason]

RescalePair looks good! I'm hesitant to approve with the formatting changes added since it is difficult to see what code is new in the Sin Cos and Tan functions. I made a separate PR to just format the code #162 . Once it is merged it should clean up this PR up a lot.

[mwoss]

Oh yea, my bad. I think Golnand formated code automatically.