Improve performance by using append to populate and grow buffer

[rykov]

Following up on PR #1, I decided to try out different strategies to replace explicit buffer management in the current version of rmarsh Generator. I tried using bytes.Buffer, bufio.Writer, and append and it looks like the latter is the best performing solution.

Here are the original performance numbers for manual buffer management in master:

BenchmarkGenReset-4         	500000000	         3.56 ns/op
BenchmarkGenNil-4           	100000000	        19.3 ns/op
BenchmarkGenBool-4          	100000000	        19.3 ns/op
BenchmarkGenFixnum-4        	100000000	        23.6 ns/op
BenchmarkGenBignum-4        	20000000	        67.1 ns/op
BenchmarkGenSymbol-4        	50000000	        30.5 ns/op
BenchmarkGenString-4        	50000000	        29.2 ns/op
BenchmarkGenFloat-4         	10000000	       222 ns/op
BenchmarkGenArray-4         	30000000	        41.1 ns/op
BenchmarkGenLargeArray-4    	    2000	   1051485 ns/op
BenchmarkGenHash-4          	20000000	        69.2 ns/op
BenchmarkGenClass-4         	50000000	        27.6 ns/op
BenchmarkGenModule-4        	50000000	        26.1 ns/op
BenchmarkGenIVar-4          	20000000	        87.6 ns/op
BenchmarkGenUserDefined-4   	50000000	        41.0 ns/op
BenchmarkGenStruct-4        	20000000	        80.4 ns/op

And here's the improved performance after switching to append:

BenchmarkGenReset-4         	300000000	         4.22 ns/op
BenchmarkGenNil-4           	100000000	        16.7 ns/op
BenchmarkGenBool-4          	100000000	        16.8 ns/op
BenchmarkGenFixnum-4        	100000000	        23.2 ns/op
BenchmarkGenBignum-4        	30000000	        48.9 ns/op
BenchmarkGenSymbol-4        	50000000	        30.7 ns/op
BenchmarkGenString-4        	50000000	        27.5 ns/op
BenchmarkGenFloat-4         	10000000	       220 ns/op
BenchmarkGenArray-4         	50000000	        36.1 ns/op
BenchmarkGenLargeArray-4    	    2000	    988043 ns/op
BenchmarkGenHash-4          	20000000	        63.7 ns/op
BenchmarkGenClass-4         	50000000	        25.2 ns/op
BenchmarkGenModule-4        	50000000	        25.7 ns/op
BenchmarkGenIVar-4          	20000000	        80.8 ns/op
BenchmarkGenUserDefined-4   	50000000	        40.0 ns/op
BenchmarkGenStruct-4        	20000000	        71.8 ns/op

The bytes.Buffer and bufio.Writer branches were much slower, so I didn't pursue those solutions. However, if you'd like to take a look, they are here: https://github.com/rykov/rmarsh/tree/buf-bytes https://github.com/rykov/rmarsh/tree/buf-bufio