refactor!: kill igraph.arpack.default and arpack_defaults lists, add arpack_defaults() as a function

[maelle]

Fix #714

@krlmlr actually I see usage of the old name in reverse dependencies, should I

• cancel the transformation of arpack_defaults to a function
• or make PRs to those packages

Whilst replacing occurrences of the old name, I noticed the opportunity to use modifyList(), so this PR does that too but in a different commit. 😇

[krlmlr]

We could allow a function in the places where arpack_default is used?

[maelle]

We could allow a function in the places where arpack_default is used?

Could you please elaborate?

In my latest commit I added a maybe too complex way to deprecate the usage of arpack_default for the options argument

• the argument is evaluated;
• if it's called "arpack_default" and is a function (it is a function from igraph, and remains a function unless the user creates an object named so), we replace it with arpack_defaults() after showing a deprecation message.

If that sounds like a good idea I'll add it (as an internal function) to other affected calls.

[krlmlr]

I don't understand this wording.

[krlmlr]

What do we need eval_tidy for? Can't we just force, and if the user has passed a function, we call it?

[maelle]

We only need to catch the case where the argument is arpack_defaults and a function. I'm using rlang::call_args(rlang::current_call())[["options"]] on the following line hence the rlang here, especially as rlang is a dependency.

Now, maybe this whole "hack" is a bad idea.

[krlmlr]

Looks good! Can you please merge when the checks pass?

[maelle]

actually no, I need to expand the hack to all functions here.

[maelle]

actually with ensure_igraph() on the main branch, I'll just close and start again.