The top line of the Git.execute docstring said that it used a
shell, which is not necessarily the case (and is not usually the
case, since the default is not to use one). This removes that
claim while keeping the top-line wording otherwise the same.

It also rephrases the description of the command parameter, in a
way that does not change its meaning but reflects the more common
practice of passing a sequence of arguments (since portable calls
that do not use a shell must do that).

These are some small clarity and consistency revisions to the
docstring of git.cmd.Git.execute that didn't specifically fit in
the narrow topics of either of the preceding two commits.

(Not specific to git.cmd.Git.execute.)

The kill_process local function defined in the Git.execute method
is a local function and not a method, but it was easy to misread as
a method for two reasons:

- Its docstring described it as a method.

- It was named with a leading underscore, as though it were a
  nonpublic method. But this name is a local variable, and local
  variables are always nonpublic (except when they are function
  parameters, in which case they are in a sense public). A leading
  underscore in a local variable name usually means the variable is
  unused in the function.

This fixes the docstring and drops the leading underscore from the
name. If this is ever extracted from the Git.execute method and
placed at class or module scope, then the name can be changed back.

Instead of swallowing GitCommandError exceptions in the helper used
by test_it_uses_shell_or_not_as_specified and
test_it_logs_if_it_uses_a_shell, this modifies the helper so it
prevents Git.execute from raising the exception in the first place.

This extracts the logic of searching log messages, and asserting
that (at least) one matches a pattern for the report of a Popen
call with a given argument, from test_it_logs_if_it_uses_a_shell
into a new nonpublic test helper method _assert_logged_for_popen.

The extracted version is modified to make it slightly more general,
and slightly more robust. This is still not extremely robust: the
notation used to log Popen calls is informal, so it wouldn't make
sense to really parse it as code. But this no longer assumes that
the representation of a value ends at a word boundary, nor that the
value is free of regular expression metacharacters.

This changes how the Popen call debug logging line shows the
informal summary of what kind of thing is being passed as the stdin
argument to Popen, showing it with stdin= rather than istream=.

The new test, with "istream" in place of "stdin", passed before the
code change in the git.cmd module, failed once "istream" was
changed to "stdin" in the test, and then, as expected, passed again
once "istream=" was changed to "stdin=" in the log.debug call in
git.cmd.Git.execute.

This is still not including all or even most of the arguments, nor
are all the logged arguments literal (nor should either of those
things likely be changed). It is just to facilitate easier
comparison of what is logged to the Popen call in the code.

In Git.execute, the stdin argument to Popen is the only one where a
compound expression (rather than a single term) is currently
passed. So having that be the same in the log message makes it
easier to understand what is going on, as well as to see how the
information shown in the log corresponds to what Popen receives.