When you have learned OpenGL 2.0, you’ll implicitly have learned many extensions that were promoted to core and just remained for backwards compatibility. Other extensions will no longer be useful because their behaviour can be done much easier with newer, more general extensions.
Have a look at the Appendix J of the OpenGL 2.0 specification. There is a list of ARB extensions. You can forget everything that was promoted to core in some version. If you’re later confronted with a lower OpenGL version, you can still quickly look up the differences between extension and core and then use the extension, there’s not much to learn there.
A few comments on the remaining extensions (perhaps a bit subjective, and I propably forgot something):
Assembly level shaders. That’s perhaps not that important, as this functionality is already covered by GLSL (core since 2.0). But as the GLSL compilers are not so good yet, they are still useful.
These two are somewhat legacy, but still important because there’s not much hardware that supports NPOT-textures (a more general extension, is included in core 2.0), so they’ll still be around for a while.
Definitely useful. That’s one of the latest extensions, basically it allows to render to textures.
I’d treat the rest of the extensions on an as-needed basis, that is, don’t learn them just because they are there, learn them only when/if you need them. The same is true for many core features, there’s much in the core that’s used rather seldom…
As for a kind of “priority list”: Don’t sort the things you learn by version number of OpenGL. For example, multitexture and texture combiners were introduced in 1.3, and then in 1.4 some restrictions of them were removed, so there’s no point in learning the restrictions.
In your place I’d jump straight to OpenGL 1.5, but don’t bother with the ARB_…_shader stuff. Make sure you really understand the whole multitexturing stuff, including texture combiners. Then learn OpenGL 2.0 (the most important change being programmable shaders).