Add support long strings

[aybabtme]

This is not perfect and I'm sure a better way could be found. It negatively affect performance on strings:

master:

BenchmarkScanNumber 50000000          58.6 ns/op    51.23 MB/s
BenchmarkScanString 20000000          98.4 ns/op   101.58 MB/s
BenchmarkScanLongString  5000000       489 ns/op   116.42 MB/s
BenchmarkScanEscapedString  10000000   149 ns/op   120.50 MB/s
BenchmarkReadString 10000000           153 ns/op    65.04 MB/s
BenchmarkReadLongString  5000000       557 ns/op   102.18 MB/s
BenchmarkReadInt  50000000            62.5 ns/op    79.99 MB/s
BenchmarkReadFloat64  10000000         163 ns/op    97.89 MB/s
BenchmarkReadBool 50000000            62.0 ns/op    64.52 MB/s
ok    github.com/benbjohnson/megajson/scanner 23.022s

This branch:

BenchmarkScanNumber 50000000          54.2 ns/op    55.36 MB/s
BenchmarkScanString 20000000           105 ns/op    94.56 MB/s
BenchmarkScanLongString  5000000       540 ns/op   105.52 MB/s
BenchmarkScanEscapedString  10000000   153 ns/op   117.46 MB/s
BenchmarkReadString 10000000           153 ns/op    65.02 MB/s
BenchmarkReadLongString  5000000       598 ns/op    95.16 MB/s
BenchmarkReadInt  50000000            64.5 ns/op    77.50 MB/s
BenchmarkReadFloat64  10000000         175 ns/op    91.14 MB/s
BenchmarkReadBool 50000000            60.6 ns/op    66.05 MB/s
ok    github.com/aybabtme/megajson/scanner  23.710s

I think this is a hack and it's not very clean. I'd like to have input as to how this could be done better. Should there be another func that is delegated to once n == len(s.scratch)?

I'll try that out next.

[benbjohnson]

Besides the append() calls it looks good to me. Sorry for the delay in reviewing it.

[aybabtme]

Before using append(dst, src...):

BenchmarkScanNumber 50000000          54.3 ns/op    55.25 MB/s
BenchmarkScanString 20000000           110 ns/op    90.27 MB/s
BenchmarkScanLongString  5000000       599 ns/op    95.15 MB/s
BenchmarkScanEscapedString  10000000   166 ns/op   108.13 MB/s
BenchmarkReadString 10000000           162 ns/op    61.69 MB/s
BenchmarkReadLongString  5000000       661 ns/op    86.11 MB/s
BenchmarkReadInt  20000000            76.9 ns/op    65.06 MB/s
BenchmarkReadFloat64  10000000         188 ns/op    84.80 MB/s
BenchmarkReadBool 50000000            63.6 ns/op    62.94 MB/s

After:

BenchmarkScanNumber 50000000          55.3 ns/op    54.27 MB/s
BenchmarkScanString 20000000           109 ns/op    91.71 MB/s
BenchmarkScanLongString  5000000       561 ns/op   101.57 MB/s
BenchmarkScanEscapedString  10000000   152 ns/op   118.31 MB/s
BenchmarkReadString 10000000           158 ns/op    62.99 MB/s
BenchmarkReadLongString  5000000       619 ns/op    92.07 MB/s
BenchmarkReadInt  50000000            65.2 ns/op    76.70 MB/s
BenchmarkReadFloat64  10000000         179 ns/op    89.03 MB/s
BenchmarkReadBool 50000000            60.5 ns/op    66.14 MB/s

[benbjohnson]

:+1:

[aybabtme]

yay!