add car-cas! and cdr-cas!

[mflatt]

Addresses #969.

[LiberalArtist]

While many procedures follow the naming convention type⁠-⁠verb, the precedents set-car! and set-cdr! led me to expect these names to be cas-car! and cas-cdr! (since they abbreviate e.g. compare-and-set-car!). Arguably, this correspondence may already be broken, since we have box-cas! despite set-box!. On the other hand, a difference is that a box is a type of value, whereas car and cdr are fields in a pair type. Chez Scheme doesn't have very many operations on boxes, but box-immutable and box-immobile both use box- as a prefix, whereas the only other entry in the index that uses car or cdr is lisp-cdr (from TSPL).

I don't have a strong opinion beyond “naming things is hard”, but I thought it was worth at least mentioning before this makes it into a release.

[JBThiel]

I suppose the -cas! naming originates from vector-set! and would best be consistent,
else confusing. The vector and bytevector have many -set!
functions, thus vector-cas!.
And box-cas! followed that, being essentially a 1-vector so you don't need an index.

I had proposed pair-cas! with an "index" designator of maybe the symbols 'car / 'cdr
to parallel the (vector-cas! vector index ...), but it's more verbose,
and not many functions take symbol args like that.

car-cas! and cdr-cas! are short and sufficiently mnemonic to me, and follow the
convention so far. Also cas as the suffix connects to the !, looks visually
distinctive.

cas- prefix is sensible too, for the reasons you mention.

I tend to think of cas! as a highly distinct operation, not really a set,
much more complicated than that. It might very well not set, and generally
requires additional logic around it. It's akin to -lock! to my sense,
which all the lock functions have as suffix.

Whether CAS prefix or suffix I'm relatively impartial, except I think it should be consistent, because a distinct conceptual domain.
So since it's -cas! suffix for the vector/box, that should probably continue,
unless wanting to re-think / rename all of them.

[mflatt]

I don't have a strong opinion, either. The other names I consider where pair-car-cas! and pair-cdr-cas! to more closely follow record - field -set!, but there aren't other functions that start pair. So, I went with the way car and cdr could be viewed as abbreviating pair-car and pair-cdr in existing function names — still keeping them at the front and cas! at the end, along the lines of the modern Scheme style and existing cas! functions.