Improve appender speed

[JAicewizard]

Rough performance before:

BenchmarkAppenderSingle/int8-16         	    8288	    142109 ns/op

Rough performance after caching chunk size:

BenchmarkAppenderSingle/int8-16         	   35270	     32326 ns/op

Full (existing API without reusing the appender for fairness) vs the new TS (TableSource) implementation

BenchmarkAppenderSingleFull/int8-16         	   11076	    109986 ns/op
BenchmarkAppenderSingleFull/int8-16         	   10000	    103503 ns/op
BenchmarkAppenderSingleFull/int8-16         	   10000	    103835 ns/op
BenchmarkAppenderSingleFull/int8-16         	   11526	    107647 ns/op
BenchmarkAppenderSingleFull/int8-16         	   10000	    107233 ns/op
BenchmarkAppenderSingleTS/int8-16           	   12663	     99329 ns/op
BenchmarkAppenderSingleTS/int8-16           	   12016	     97747 ns/op
BenchmarkAppenderSingleTS/int8-16           	   12054	     98896 ns/op
BenchmarkAppenderSingleTS/int8-16           	   12721	     96048 ns/op
BenchmarkAppenderSingleTS/int8-16           	   12434	    102425 ns/op

This benchmark only tests the single threaded row based tablesource, if this gets looked at and there are no major problems I can implement the multi-threaded and chunk APIs as well. These will be even faster!

[taniabogatsch]

Thanks!