Enable multiple function composition in prefix form

[JeffFessler]

Document the prefix form and extend the prefix form to support composition of 3 or more functions.
If there is favorable sentiment for this then I can also add tests.
There was discussion in 2016 about related issues but I don't see anything more recent
#17184

[dkarrasch]

Since this is a new feature, I guess we will need to announce it in a one-liner in NEWS?

[StefanKarpinski]

Nice addition! Makes perfect sense that composition should work for arbitrary args.

[StefanKarpinski]

Looking good. Next round of improvements:

1. A couple of simple tests (although there's already the doc test)
2. Add a !!!compat "Julia 1.4" annotation in the docstring indicating that this is a new feature
3. Add an entry in NEWS linking to this PR and briefly describing the feature.

resolved

[JeffFessler]

I've done the "next round of improvements" list. Just made some last tweaks...

[dkarrasch]

A bit off-topic: @StefanKarpinski I saw you dismissed my earlier review, which was referring to a much earlier state of the PR. Should I have actively changed my review assessment as to not confuse people? Obviously, it was not meant to "request changes" indefinitely. Can I take back a "request changes" without necessarily "approving" the PR?

[fredrikekre]

This PR failed CI due to trailing whitespace. Edit: #33574

[StefanKarpinski]

Hmm. I'm not sure. I think you can mark individual comments as resolved but I'm not sure how you mark a review as resolved other than what I did — i.e. dismissing it, which felt a bit, uh, dismissive, but seemed like the only thing to do in order to let the PR be mergeable. Maybe leaving a "comment" review puts it into a neither approved nor denied state?

[StefanKarpinski]

It's really annoying that GitHub will show you the deletion of trailing whitespace in a diff but won't show you the insertion of trailing whitespace, so it's impossible to tell by reviewing a diff that it introduces trailing whitespace. Sorry I missed the whitespace failure in CI. I'm so used to a lone failing CI test for simple changes like being inconsequential that I just ignored it out of habit.

[JeffFessler]

Thanks everyone - one line of new code and two dozen lines of comments and tests :)

[StefanKarpinski]

That's the way it goes 😁

[jw3126]

I think we should also have

∘() = identity
∘(f) = f

[StefanKarpinski]

Seems like a good addition. Care to make a PR, @jw3126?

[jw3126]

Sure thing!