Parse doubles with boost::spirit::qi::long_double

[jimhester]

This works around an unaddressed spirit regression
(https://svn.boost.org/trac/boost/ticket/11608).

Fixes #412

I did some benchmarks and there was no performance regressions when parsing a large file of doubles.

Note the numbers being tested are both representable as regular doubles, the double_ parser is just parsing them improperly.

[hadley]

LGTM

[defconst]

real_parser<double, DecimalCommaPolicy> needs also a fix, it has the same problem with large numbers ...

reprex: https://gist.github.com/stdlib/75d90e9bbde3762eb78acc8f5271232d

> ./real_parser 10000000000000000000,00
real_parser<double>
return value: 1
parse result: 1e+19
real_parser<long double>
return value: 1
parse result: 1e+19
> ./real_parser 100000000000000000000,00
real_parser<double>
return value: 0
parse result: -1
real_parser<long double>
return value: 1
parse result: 1e+20

[jimhester]

@StdLib thank you for noticing that, now fixed as well.