First pass at adding constexpr where useful

[oxt3479]

This is a first pass through Imath to add constexpr to functions that are contenders for compiler optimization.

[lgritz]

What's here is certainly LGTM and acceptable (and very welcome!), and as far as I'm concerned, we should merge it even if we add more constexpr later.

But what I can't tell from these diffs is the degree to which this job is complete.

Are there methods of these classes that you were not able to constexpr? Can you enumerate or summarize which those were?

Also note that the rules for what can be constexpr changed/expanded a bit from C++11 to 14 to 17. I assume that this PR is for the "least common denominator" -- using C++11 rules. If there are yet other things that can't be constexpr under C++11 but can be for 14 or 17, we should at least note them in comments, add a "long-term issue" to remind us to revisit once our supported minimum is bumped to C++14, or consider using a macro such as these that I've found helpful: https://github.com/OpenImageIO/oiio/blob/master/src/include/OpenImageIO/platform.h#L77

[oxt3479]

Concerning completeness: Every file within the Imath/Imath subdirectory has constexpr added wherever possible in accordance with the c++14 standard. The is except for methods which return void, as these will only be 'fruitful' when they return an error which did not seem worth any added compile time.

Discretion regarding c++11 was overlooked but is being addressed in a separate pull request using a macro to differentiate.