Add 1+, 1- and -1+, and improve abs in cptypes

[gus-massa]

Two slightly relate patchs.

Add 1+, 1- and -1+ to cptypes

I should have added them in the (old) previous commit that added add1 and sub1, but I forget these primitives existed. I refactored some of the test to avoid too much duplication.

Improve fixnum case of abs in cptypes

In the (old) previous commit I added a reduction for (abs <flnumber>) to (flabs <flnumber>). But it has no reduction for (abs <fxnumber>).

Now I'm reducing (abs <fx>) to

(let ([t <fx>])
  (if (fx= t (most-negative-fixnum))
      (pariah (- (most-negative-fixnum)))
      (fxabs t)))

that is the straightforward expansion. It's expanded later in cpprim to

(let ([t <fx>])
  (if (fx= t (most-negative-fixnum))
      (- (most-negative-fixnum))
      (if (fx>= t 0) t (fx- t))))

But perhaps it's better to expand it in cptypes directly to

(let ([t <fx>])
  (if (fx>= t 0)
      t
      (if (fx= t (most-negative-fixnum))
          (- (most-negative-fixnum))
          (fx- t))))

becuse it is faster for positive numbers and slower for (most-negative-fixnum) that is very unusual.

I also wanted to add a similar special case for add1 and sub1, but in their expansion in cpprim they use the overflow flag, so expanding them to a comparison to (most-negative-fixnum) or (most-positive-fixnum) is probably slower. (I hope to try something like this next month.)

[burgerrg]

Thank you for the updates! Please add a subsection to the release notes describing them.

[gus-massa]

I fixed one misplace parenthesis that was causing an error in one of the tests.

I added a note for each commit. It is it fine or it's better to accumulate all the changes in a single note?

But I did't add this text in the notes to avoid repetition. I can add it if you think it's necessary.

Type recovery is enabled with the \scheme{enable-type-recovery} parameter.

[burgerrg]

I'd combine these into a single entry, something like:

\subsection{Type recovery improvements (10.2.0)}

The type recovery pass has improved support for \scheme{abs} with a fixnum argument
and added support for \scheme{1+}, \scheme{1-}, and \scheme{-1+}.

[burgerrg]

It's fine not to add the sentence about enabling type recovery.

[gus-massa]

Done

[mflatt]

Should this test have been kept? I don't think it's covered by the test-closed1 generalization, but I might be missing something else.

[gus-massa]

I undeleted the test. It's a to ensure I don't make an easy mistake in the future. I don't know why I deleted it.

It's not included in test-closed1. Moreover, I was thinking about writing test-not-closed1 to reduce duplication, but I decided that it was too weird. But Perhaps I'll add it in the future if there are too many cases of almost-closed functions.