bug#57129: 29.0.50; Improve behavior of conditionals in Eshell

[Jim Porter]

Here are some patches to fix this.

The first patch adds tests/documentation for the current state so that 
the subsequent patches are clearer. It also improves 
'eshell-close-handles' so that you can set the exit status elsewhere, 
which makes some of the code simpler.

The second patch fixes the use of variables in conditionals (e.g. if 
statements, while loops). The only non-test code change here is that 
'eshell-structure-basic-command' needed to be taught that forms 
beginning with 'eshell-escape-arg' should be treated as data, much like 
'eshell-convert'.

The third patch fixes the behavior of the '$?' and '$$' variables to 
match the manual: '$$' is t when an external command succeeds, and '$?' 
is 1 when a Lisp command signals an error. I also added a new feature 
that '$?' is 2 when an actual Lisp *form* returns nil. This lets you use 
Lisp forms as conditionals in Eshell much like you would in regular 
Lisp. However, it doesn't apply to the command syntax, even if it runs 
Lisp code:

  ~ $ (zerop 1); echo $?
  2
  ~ $ zerop 1; echo $?
  0

That's because Eshell prints the return value of Lisp commands (unless 
it's nil), so a Lisp command that wants to silently succeed would return 
nil. For example, the Lisp version of 'cat' returns nil. We want that to 
have an exit status of 0. However, when writing it as a Lisp form with 
parentheses, I think it makes more sense that nil is treated as false 
for conditionals. The overall goal is so that this prints hi:

  cat foo.txt && echo hi

And that this doesn't:

  (zerop 1) && echo hi

For people who don't like this behavior, they can customize the new 
'eshell-lisp-form-nil-is-failure' variable.

0001-Only-set-Eshell-execution-result-metavariables-when-.patch
Description: Text document

0002-Allow-using-dollar-expansions-in-Eshell-conditionals.patch
Description: Text document

0003-Make-and-variables-more-consistent-in-Eshell.patch
Description: Text document