All unknown (and, formerly, silently ignored) keywords were removed from
existing tests. This allows us to detect typos.

Default OPCODE is QUERY so existing tests should not require changes.

refs: #11

Clearing and writing to TIMESTAMP_FILE is not atomic and libfaketime
can read empty file which represents default behavior and returning
realtime. Create and move new file is atomic and prevent this issue.

RFC 5011 implementation in kresd and Unbound require separate
trust anchor file for each domain.

e.g. provide real information about step in question

Deckard now can run multiple processes inside single simulated network
and communicate with each other.

This introduces couple of incompatible changes:
- command line syntax was changed to accomodate new possibilities (see --help)
- SOCKET_WRAPPER environment variables except DIR are always managed by Deckard

Further work requires importing pydnstest.scenario into testserver.
Socket functions were moved into scenario to avoid cyclic imports
between testserver and scenario modules.

We should be moving the socket functions in the other way,
from scenario into testserver. This will be subject of later
refactoring.

This is handy for simulating broken servers which are 0x20 sensitive.

Previously invalid use of raw_cmp() would generate an exception which
would be incorrectly caught up in the call tree.

Socket family selectino logic was unnecessarily duplicated several times.

Trivial changes to pythonify boolean expressions and avoid pep8 error E712
"comparison to True should ..."

Cheap re-indentation using python-autopep8-1.2.1-3.fc25 with few manual
tweaks for very long lines.

This costs nothing and will avoid PEP8 complaints about whitespace in CI.

Cheap re-indentation using reindent.py from python-2.7.13.
This costs nothing and will avoid PEP8 complaints about tabs in CI.

The main problem was that socket.type from Python standard library is
not an integer anymore. It is EnumInt object which is used instead of
bit field. To make the code compatible with both versions of Python we
had to replace all exact value comparisons with bitwise masking.

Custom implementation of packet logging was replaced by
SOCKET_WRAPPER_PCAP_FILE facilities. The main advantage is that it logs
all the traffic including malformed packets, ICMP messages and so on.

As side-effect, this commit removes depedency on Python package dpkt.

dnspython library by default coalesces RRsets of same type into one and
automaticaly deduplicates equicalent RRs inside one RRset.
This prevented Deckard from testing DNS messages with duplicate RR.

To prevent this, all DNS messages are now parsed using option
one_rr_per_rrset=True. This setting may create multiple RRsets
of the same type but it seems okay for Deckard because the comparison
algorithms compare individual RRs and ignore how they are distributed
among RRsets.

This is necessary to avoid
Instance of 'Step' has no 'queries' member (no-member)
errors reported by Pylint.

For easier debugging we will need proper logging with categogies and
levels. For now just cleanup redundant if VERBOSE: tests around
dprint() calls.

Multiple IPv4/IPv6 addresses can be specified for single RANGE.
This allows to specify responses for zones with multiple
name servers without copy-pasting the data for each server's address.