This patch fixes a group of failures in gdb.base/shell.exp on Windows 
host due to assumptions that GDB is launching a POSIX-like shell.  On 
Windows, we get CMD.EXE instead.

There is a potential second group of failures on Windows host not 
addressed by this patch, due to the dependence of some of these tests on 
utilities like wc, sed, and grep.  These tests happen to work for me 
because the test harness connects to the remote host using Cygwin ssh 
and both GDB and CMD.EXE inherit the PATH setting from the parent Cygwin 
shell.  I don't know if disabling those tests on Windows host too is 
appropriate, or again if this is an instance of badly-designed tests 
that could be rewritten to avoid those dependencies.  WDYT?

-Sandra