The multiplicity of ways to do everything in OpenGL was always a stumbling block for me in the past; it's been a big driver towards leaning in the direction of DirectX, or Monogame or Unity for cross-platform, in the past.
There still seems to be a lot of fragmentation from the handful of samples I've looked at so far; is there something like a canonical example of "the right way" to do things?
Yeah, first thing you have to do is dynamically collect extensions on program startup.
https://vulkan-tutorial.com/Drawing_a_triangle/Setup/Instanc...
That's a bit dishonest way to put it. The situation with extensions has vastly improved w.r.t. OpenGL.
In the link you provided, it only needs to enable the appropriate window system extension. It also prints out some info about other extensions but that is not required.
It's a huge improvement that extensions need to be explicitly requested at init time.
It's massively better than OpenGL, and API extensions can't be completely avoided in a multi-vendor cross-platform API without limiting to lowest common denominator.
It doesn't look much different from OpenGL multiple execution paths for each GPU, and possible workarounds when they aren't available.
Specially given the amount of extensions that have been added since Vulkan was released.
OpenGL ES 2.0 and higher cleans a lot of that up. Crufty old calls are removed. You simply include a header and link a library without the need for a complex loader (though there still are extensions).
Vulkan is a much better designed API with a lot less legacy crap in it. It's lower level and gives more control to the app/engine developer.
The bad news is that Vulkan is a bit verbose, but this stops being an issue once you move past "hello world" level apps. OpenGL gets really difficult once you start doing complex things.