Make `abs` work for complexes

[jpellegrini]

R7RS says it "computes the absolute value of its argument", but doesn't imply it should be real... So we do compute the norm, (abs a+bi) $= \sqrt{a^2+b^2}$ as usual.

A very small patch... :)

(abs -3+4i)             => 5
(abs -3.0-4i)           => 5.0

Here's a list of how other lisps deal with (abs -3-4i):

System	result	remark
Bigloo	no complexes
Biwa	no complexes
Chez	error
Chibi	-3-4i	leaves unchanged
Chicken	error
Cyclone	error
Gambit	error
Gauche	5.0
Guile	error
Kawa	5.0
LIPS	error
MIT	error
Racket	error
Sagittarius	5	keeps exactness
STklos	error
Unsyntax	-3-4i	leaves unchanged
ABCL	5.0
CCL	5	keeps exactness
Clisp	5	keeps exactness
ECL	5.0
GCL	5.0
SBCL	5.0

[lassik]

R7RS says the signature is (abs x). For the number procedures, x means a real number. z ("zahl") means any number.

Nevertheless, R7RS doesn't forbid procedures that work with more data types than is standard.

[jpellegrini]

R7RS says the signature is (abs x). For the number procedures, x means a real number. z ("zahl") means any number.

I haven't paid attention to the signature... Indeed! Anyway, @egallesio, here's the PR. Any decision seems to be fine! :)

[jpellegrini]

Hm, I think I hadn't had coffee when I proposed this PR. If one wants to make abs work for complexes, it would be simpler to do something like

(define abs magnitude)

No?

[egallesio]

I have applied your patch (and simplified the writing of magnitude, BTW)