Rework the test suite

[krizhanovsky]

There are several architectural issues in the current test suite (the list is going to be updated):

• There are 2 versions of the framework: the whole test suite is the old one, while /framework is the new one. Need to finish moving to the new framework.
• The test framework must use HTTP/2 client connections by default. HTTP/2 has several drawbacks, performance and observability, so it actually doesn't replace HTTP/1 in all the cases (e.g. server management) and we should not throw HTTP/1 out.
• /framework and /helpers separation is unclear: helpers are expected to contain generic classes, e.g. dmesg or analyzer, useful in different tests directly, while deproxy.Client seems purely dedicated to the framework itself and it's strange to see deproxy.py in /helpers while all the related logic is in /framework.
• Master branch of tests is not compatible with older releases. There are two sources of the problem: new features and changes in the configuration file format. It'd be better to have one test code base for all the versions. New features can be disabled in tests_disabled.json with a special version field, e.g. "version" : "0.7" means that the test is disabled for Tempesta version below 0.7. In the new framework, the full Tempesta configuration is listed in the test description file, not compiled from parts, so we can give very specific configuration for exact Tempesta version. Some good and generic solution is required. (Duplicate of Tests for stable versions #55)
• dmesg is analysed after every test case to search for possible oopses and warnings. Current way of analysis (dmesg | tac | grep -m 1 "Start test" -B %i | tac) is not effective for long running tests, so dmesg will be called once again. Instead we can simply use journalctl -k --since=${date} to show all required content with one call. More over, journalctl can be configured to use really big buffers, and full system log will be available at any time.
• Move to Python3 since Python 2 in supported only until 2020. The only external module which prevented us from migration to the 3rd version was Scapy-TLS, which has been removed in Moving away from Scapy-ssl_tls and fix tls/ tests #179 . 2to3 can be used for the migration. https://github.com/horlabs/scapy-ssl_tls seems made some iteration to port the library to Python 3. Please update https://github.com/tempesta-tech/tempesta-test/blob/master/README.md when this point is done.
• The last commit in Scapy-TLS was in Sptember 2018, so we copied its file to the test suite . It seems the base Scapy supports TLS. Scapy-TLS also has bugs like UserWarning: Verification of GCM tag failed: MAC check failed tintinweb/scapy-ssl_tls#163 , which affect us, e.g. in Verification of GCM tag failed: MAC check failed tempesta#1310 .

The both the current test frameworks use HTTP/1.1 in plain test mode, w/o TLS, while the most real life traffic is at least encrypted and HTTP/2 grows. It means that out test suite tests the most crucial components in only ad hoc tests. That doesn't make much sense, so I mark the issue as crucial.