switch effect analysis to snake case

[LilithHafner]

This PR was formed by running a bunch of global find-and-replaces

find . -name *.jl -exec sed -i 's/inaccessiblememonly/inaccessible_mem_only/g' {} \;

And manually updating the public @assume_effects macro to continue accepting the old symbols as arguments.

From discussion on Zulip (https://julialang.zulipchat.com/#narrow/stream/137791-general/topic/About.2Ejl/near/436262421)

This seems worth doing in the light of the inconsistency between inaccessiblememonly and effect_free and the fact that effect analysis is slowly becoming public.

Very open to adjusting the scope of this PR/how much we rename. I leaned pretty far into making all the effect analysis snake case in this first draft to see what folks think and to avoid having to make followups.

[non-Jedi]

Particularly of note (for motivating this change) from the linked zulip discussion is the fact that somebody read inaccessiblememonly as in_accessible_mem_only and thus reversed the meaning.

[christiangnrd]

Should :noub_if_noinbounds be changed to no_ub_if_no_inbounds?

[Keno]

Not a fan of separating nothrow and noub. They are one word in my mind, but then again, I am German. effect_free is probably just badly named in general, because @assume_effects implies a much broader notion of the word effect. Perhaps unobservable would be better. inaccessiblememonly was named because that's the LLVM name for this, but we could perhaps do better.

[mikmoore]

Maybe nonaccessiblememonly or nonaccessible_mem_only? But inaccessible_mem_only also seems fine and is more grammatically conventional (I think).

[tecosaur]

It seems as though this PR has been a bit forgotten about?