This way, we can effectively kill all subprocesses. This worked for
slapproxy because Flask calls setsid internally, but slapgrid-* doesn't, so
they wouldn't be killed by killpg.

Fixes upgrade path from older erp5testnode version.

- display the output of all commands launched by erp5testnode in
  erp5testnode log
- add a thread to upload ongoing logs to the master regularly
- if the software release is not built successfully after a
  few time, totally erase software. This help unblocking if
  buildout is unable to update software.
- check if the last test result was cancelled in order to
  allow relaunching test without restarting testnode
- improve logs to allow much better understanding of what is
  going on

Since the remote was not cleaned when the runTestSuite was totally
failing, there was no chance to get a higher revision with potentially
a fix. Now errors will be reported to the master.

data (ex. MySQL db content) from previous testnode's runs

by trying reconnect before server keep-alive ends and the socket closes.

restart if task distribution node is down.
Refactor.

so be tolerant and run it a few times before giving up.

as partitions can be of any kind we have no reliable way to check if they are up or not ...

But Still store the use-case elapsed time as integer.

- add no header section to define the header is not existing depending on
  the condition