Add support for enum <--> string conversion

[daljit46]

There are a lot of parts of our codebase, especially the CLI, that rely on strings for configuration and option handling, where using enums would be much nicer.
C++ unfortunately (and astoundingly) doesn't have builtin support for enum <--> string conversion. C++26 added reflection, which will finally allows to build such a basic utility but it'll be a few years until our codebase can make use of that.
In #3258, I made use of magic_enum for this purpose. It works very well and I think we should consider adopting this, perhaps wrapping the library in our own interface. An alternative is enchantum, which has the added benefit of lower compilation times due to template instantitations (which is something we should consider given that our compilation times are not exactly excellent).
AFAIK, both libraries are relying on compiler specific behaviour to achieve this, but all major compilers (GCC, Clang and MSVC) are supported.

[daljit46]

An alternative of course is using macros and while I am not exactly a fan of them, I think this would a justifiable usecase and avoid requiring a new dependency to be added to the project.

[daljit46]

@MRtrix3/mrtrix3-devs anyone has opinions on macros vs third-party lib here?

[Lestropie]

+1 for increased enum usage in general. The more I've invested in code quality changes the more I realise the scope for errors when passing bools around, especially between functions.

One of the main applications for enums that necessitates string representations is the CLI. So any solution would want to expedite both string-to-enum conversion (ideally case-insensitive) and construction of the set of strings of "choices" of which to make the CLI aware so that it can be non-templated.

I would probably lean toward magic_enum as the more established and popular. I would not have thought that templated enumerations would contribute little to compilation time compared to the rest of the back-end. One more external dependency is I think not a major problem given it would be cmake-managed, ie. anyone compiling from source wouldn't even need to install an additional system package.

[daljit46]

Ok I will try to make a PR with magic_enum soon and maybe it's a good idea to benchmark the compilation times too (although as you said it may be the case there is no noticeable difference).