Fix variable evaluation on 'evaluate-stream'

[Affonso-Gui]

Splitting from #362

Targeting cl-compatible since this needs some testing before practical use.

Below is copy of the description.

Added commit to solve the following problems in toplevel variable evaluation:

• Cannot evaluate a symbol that is also fbound
• Evaluation of unbound symbols leads to unclear response: 32512

With this changes:

1.eus$ 10
10
2.eus$ *
10
3.eus$ * 2 2
4
4.eus$ ls
c               Makefile.Cygwin    Makefile.IRIX5      Makefile.LinuxSH4.2    Makefile.SunOS5.X11R6.1
comp            Makefile.Cygwin32  Makefile.IRIX6      Makefile.Linux.thread  opengl
geo             Makefile.Cygwin64  Makefile.Linux      Makefile.rgc           README
image           Makefile.Darwin    Makefile.Linux64    Makefile.SunOS4        README.rgc
l               Makefile.generic1  Makefile.LinuxARM   Makefile.SunOS4.sub    tool
Makefile        Makefile.generic2  Makefile.Linux.ppc  Makefile.SunOS5        xwindow
Makefile.Alpha  Makefile.IRIX      Makefile.LinuxSH4   Makefile.SunOS5.i386
0
5.eus$ guiga
sh: 1: guiga: not found
Call Stack (max depth: 20):
  0: at #<compiled-code #X52d7e80>
eus 0 error: unbound variable guiga in #<compiled-code #X52d7e80>
6.E1-eus$ 

Note that

1. * is symbol-evaluated when there are no trailing arguments and function-evaluated otherwise
2. unbound variable errors are appropriately raised on evaluation of unbound symbols
3. unix evaluation remains the same for valid commands

[Affonso-Gui]

Maybe in some later releases (too dangerous for now)