units: Improve precision of angle conversion

[ghost]

Use high precision constants from boost/math/constants/constants.hpp
to fix trac#6893.

[mkurdej]

Looks good to me, but I don't know what's the policy about dependencies on other Boost libraries. Anyone?
Personally, I find it ok, as I would probably use math constants together with units anyway, so there will be no overhead actually.

[jhunold]

We already depend on Boost.Math, see units/cmath.hpp for the functions used there. But I'd like to have a unit test for this. My main concern is the question of hardcoding double precision instead of using whatever precision the user wants.

[ghost]

@jhunold But the double is hardcoded here:
https://github.com/boostorg/units/blob/master/include/boost/units/conversion.hpp#L167
Unless I'm mistaken, the user doesn't have a choice here.

I opened https://svn.boost.org/trac/boost/ticket/11389 about this issue.

[jhunold]

@rleigh-dundee the original double is okay as it is a POD and is subject to the usual decaying rules for PODs. Using a double as a template parameter triggers template instantiation rules which opens up a different can of worms . So the hardcoding of double is not a separate issue, but a part of the original issue #6893. Those are the reasons I want an unit test case which shows that different constants are used for different underlying floating types and makes sure that this stays so.

[hernando]

Is #44 a valid solution? The current constants do not have acceptable precision when using double.