Fix error reporting in EC2 auto discovery

[marcoandredinis]

To auto discover EC2 instances, the Discovery Service calls ssm.SendCommand to multiple instances.

Then, it will wait for the command to be executed.
The waiter will check for the output of the execution and:

• return an error when the command execution takes too long (waiter waits up to 100s)
• return an error when it fails to check for status (eg, network issues, permission issues, ...)
• return an error when the command fails

The last case was not being handled gracefully.
So, the Discovery Service would assume it's an unknown error and bubble up.
This meant not error reporting: neither in logs, nor in audit log.

This PR changes the flow, so that command failures are handled gracefully.
This will ensure the audit log gets an entry which identifies the problem.

changelog: Fix EC2 auto discovery report of failed installations.

[backport-bot-workflows]

@marcoandredinis See the table below for backport results.

Branch	Result
branch/v18	Create PR