Fix ENIP test crashes and conpot_protocol snafu

[srenfo]

This PR fixes multiple issues that lead to the ENIP tests crashing without much notice. The root cause was a major bug in the conpot_protocol decorator. It also adds a few tests to catch regressions.

Another related issue is that Conpot (in both production code and test code) generally does not properly shut down greenlets. A general fix for that proved to be too big for this PR.

1. The ENIP test setUp() method was crashing, but the exceptions weren't propagated from the greenlets, thus going unnoticed. There's even a comment that acknowledges that there's something wrong:

   # TODO: udp does not cleanly shutdown. We get OSError.

   The first commit exposes the crashes but does not fix anything just yet.

2. The second commit fixes conpot_protocol allowing only a single instance reference per class:

   core_interface.protocols[self.cls] = self.wrapped
   [...]
   return getattr(core_interface.protocols[self.cls], name)
   

   This effectively turned every decorated class into a singleton, where earlier instances would share instance variables with the most recently created instance. This manifested itself in the ENIP tests because setUp creates instance for TCP and UDP, respectively.

3. The third commit fixes ENIP greenlets surviving across test runs. The UDP ENIP servers were never shut down properly. This is the commit I'm least happy with because the check in handle_udp feels somewhat hacky. But it mirrors the pre-existing check in handle_tcp, except it breaks out of the function entirely. The tests now wait for the greenlets to be done. With these three fixes, the tests run and succeed without crashing. The server.stop()/greenlet.join() pattern in tearDown() should be repeated throughout the code, but as mentioned above a proper refactoring was too big for this PR. You don't want to copy&paste it everywhere because it's bound to be forgotten again. Some tests already have similar but not quite identical code.

4. The fourth commit improves ENIP test runtime and robustness: Tests were either hitting a timeout on success or reporting success anyway. That's why the ENIP test suite took >=12 seconds. It now takes just over 1 second on my machine AND response errors are no longer masked.

[coveralls]

Pull Request Test Coverage Report for Build 1183

• 14 of 17 (82.35%) changed or added relevant lines in 2 files are covered.
• 38 unchanged lines in 4 files lost coverage.
• Overall coverage increased (+0.8%) to 73.923%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
conpot/protocols/enip/enip_server.py	9	12	75.0%
<!--	Total:	14	17	82.35%	-->

Files with Coverage Reduction	New Missed Lines	%
conpot/protocols/http/command_responder.py	6	66.46%
conpot/protocols/enip/enip_server.py	7	71.63%
conpot/protocols/ftp/ftp_handler.py	11	82.23%
conpot/protocols/ftp/ftp_base_handler.py	14	81.35%
<!--	Total:	38		-->

Totals
Change from base Build 1181:	0.8%
Covered Lines:	5610
Relevant Lines:	7589

---

💛 - Coveralls

[creolis]

Thank you for your contribution :)

LGTM @xandfury?

[xandfury]

LGTM! Thanks for your cotribution :slightly_smiling_face: @srenfo cc: @creolis