Better unit tests for bash/zsh completion scripts

[remkop]

The current unit tests for AutoCompletion test by asserting that the generated script equals some pre-defined script; they don't verify that completion actually works in bash or zsh.

Ideally, the tests would verify that a generated completion script actually generates the expected completion candidates, given some partial input.

Useful links:

• https://stackoverflow.com/questions/9137245/unit-test-for-bash-completion-script
• https://stackoverflow.com/questions/3520936/accessing-bash-completions-for-specific-commands-programmatically
• https://iridakos.com/programming/2018/03/01/bash-programmable-completion-tutorial
• https://echorand.me/posts/linux_shell_autocompletion/
• https://stackoverflow.com/questions/22795767/how-to-test-python-readline-completion
• https://debian-administration.org/article/An_introduction_to_bash_completion_part_1
• https://debian-administration.org/article/An_introduction_to_bash_completion_part_2

(Related to #894 but different)

[remkop]

Work is in progress to create a test suite for interactive tests using the DejaGnu test framework (see also the DejaGnu wikipedia page). DejaGnu is based on Expect (see also the Expect wikipedia page and the Expect man page. Expect in turn is written in TCL.

I will be borrowing quite a lot of code from the bash-completion project, even though that project has started to move its testing to pexpect (a python-based implementation of expect) for unknown reasons.