[3006.x] Fix salt-ssh retcode reporting

[lkubb]

What does this PR do?

• Ensures invalid retcodes reported by the remote do not crash salt-ssh.
• Also generally ensures salt-ssh does not crash if retcode is a non-integer for whatever reason.
• Makes sure salt.client.ssh.shell.Shell.exec_cmd calls term.close before querying the exit status and returns an exit status > 128 if a process died as the result of a signal. Both changes avoid returning None as retcode.

While I could fairly reliably reproduce the reported behavior instrumentalizing another PR for running the test suite many times in parallel, the changes to Shell found here completely solved the intermittent test failures in two full runs, where previously one could find 3-4 instances per full run (which usually sorted out themselves during the re-run and thus were not spotted; full run meaning all test runs that are part of the CI pipeline for PRs, not a single full run of the suite on a system).

What issues does this PR fix or reference?

Fixes: #64575
Fixes: #64588

Previous Behavior

Seldomly: TypeError: '>' not supported between instances of 'NoneType' and 'int'

New Behavior

Exits with the correct exit code and returns output as usual.

Merge requirements satisfied?cs

• Changelog - https://docs.saltproject.io/en/master/topics/development/changelog.html
• Tests written/updated

Commits signed with GPG?

Yes

[lkubb]

@dwoz Friendly ping. :)

General remark: This patch has been carried by QubesOS since June to fix their relatively frequent CI failures caused by the underlying issue.