slapgrid: Better detect when master is unavailable

Improve master availability detection for online/offline instance processing: if instance list is successfully fetched from master, proceed, otherwise whatever the reason, fallback to just starting the services of existing instance.

Note that connection errors can occur in later steps as well, and that the instance list is susceptible to be cached if there are several runs within the same Python process.

Adapt tests, because /getFullComputerInformation is cached within the lifetime of a Python process, so the tests were not exactly mimicking consecutive runs.