Add support for $INCLUDE directives in zone files

See merge request dns-testing-tools/ibdns!39

The same passage also stated that only a small subset of resource record
types was supported by the parser. That is no longer true, and most zone
files can be read in without a problem.

It works, but there is no protection against inclusion loops yet.

Make data structures describing errors happening at zone loading time
more consistent. The idea is to always ensure that error tuples are
three-tuples of {:error, where, what}.

The parser code is also improved so that errors happening at
post-processing time stop that post-processing immediately, instead of
waiting until the file is done processing.

Prepare test zones that contain $INCLUDE directives. The unit tests are
only modified so that they read from the files instead of hard-coded
strings.

Clarify that the origin parameter in IBDNS.Zone.FileParser.read/3 must
be an domain in absolute labels list form, be it in the documentation
and in the spec.

To enforce that restriction, also define a useful guard in IBDNS.Name
and use it for IBDNS.Zone.FileParser.read/3 so that anything that is not
an absolute labels list is rejected.

Relicense project as GPLv3

See merge request dns-testing-tools/ibdns!38

Add or improve docstrings in various modules, types and functions, to
make the source easier to follow for newcomers.

Some functions have specs added to them. This improved the coverage and
exhaustivity of Dialyzer reports and helped fix one bug.

There are some other minor drive-by refactorings in this code, but they
shouldn’t change anything functionally.

Ensure compliance to the REUSE specification by adding another action in
the testing stage of the continuous integration setup.

After stating the license in a form that’s nice for machines, let’s
state it in a more human-friendly and prominent way by editing
README.md.

Mark all files in version control with the appropriate licensing
information, according to <https://reuse.software>.

All files are licensed under GPL-3.0-or-later, with some exceptions.
There are two CSV files downloaded from IANA whose copyright information
is set appropriately (CC0-1.0). Some trivial files are released under
the CC0-1.0 license too.

Don’t crash if a virtual server has no zones

See merge request dns-testing-tools/ibdns!37

Fix a crash in IBDNS.Server.Catalog that could occur if a query was sent
to a virtual server for which no zone is loaded.

TCP listener: avoid buildup of DOWN messages

See merge request dns-testing-tools/ibdns!36

TCP listeners run separate acceptor processes that they monitor. In the
happy path where a client connects, the acceptor process sends a message
to the parent, then exits.

That happy path causes two messages to be sent to the parent process: a
tcp_connect message and a DOWN message. But the DOWN message was never
handled, causing the message queue of the TCP listener to build up and
grow unbounded.

The fix consists of demonitoring the process that just gave the TCP
listener a socket. For good measure, we give the :flush option, which
removes a possible DOWN message from the caller’s message queue. This
should be enough to fix the message queue growth problem.

Ideally, this entire module could use an overhaul. The whole thing is
hacky because of the need to call :procket.open and :procket.close/1
manually when starting or stopping a listener. There ought to be a
smarter way.

Update csv and dialyxir; improve CI by caching

See merge request dns-testing-tools/ibdns!35

There are some redundant steps in the build and test scripts that cause
dependencies to be downloaded. Leverage caching so that we don’t do this
too often.

The source code repository has moved, so update the @source_url
attribute in mix.exs.

Also, Elixir 1.16+ has started requiring that yecc be explicitly listed
as a necessary compiler if the project needs it.

Update versions of build-time, development and test dependencies.

The csv package bumps major versions but the changes do not affect us.

As for dialyxir, the later minor version fixes a potential problem with
more recent versions of Elixir (1.15 and up).

Replace ~R with ~r in unit test

See merge request dns-testing-tools/ibdns!34

There was still one place where ~R was used, but without causing
continuous integration to fail. Let’s tackle that one as well.

Discontinue use of ~R sigil (part 2)

See merge request dns-testing-tools/ibdns!33

Really remove all uses of the ~R sigil in the code, in places where a
previous attempt at removal of this sigil have missed.

This time, the replacements were made in actual code, not in the project
definition. Unit tests should pass.

Discontinue use of ~R sigil

See merge request dns-testing-tools/ibdns!32

Remove all uses of the ~R sigil in the code because it is deprecated,
and removed in Elixir 1.16.

The only places where these ~R regular expressions were used can be
easily replaced with the ~r sigil.

Manually building the documentation with “mix docs” shows no adverse
side-effect by this change.

README: we aren’t alpha software anymore

See merge request dns-testing-tools/ibdns!31

Release 0.3.3

This release addresses a design flaw regarding the handling of data
stored in Catalog structures at keys differing only by case. This
resulted in some resource records being only partially served.

Correct handling of zone files containing nodes with inconsistent case

See merge request labs/ibdns!30

This commit fixes an oversight leading to the server not responding as
intended when loading a zone file such as this one:

  $ORIGIN bad.example.
  $TTL 3600

  @               IN SOA […]

  alias           IN CNAME target1
  ALIAS           IN CNAME target2

Querying alias.bad.example should yield two CNAME resource records. In
its default configuration, the server should have responded with
SERVFAIL, but didn’t.

This commit fixes a subtle, yet embarrassing bug that stemmed from
IBDNS.Catalog formerly failing to specify in a satisfactory way what
happens if someone inserts a datum at a key, then another datum at a key
that differs from the first one by case only.

Because of that, there was an issue when the server loads a zone file
where nodes are not written with consistent case. Consider the following
example:

  $TTL 3600
  $ORIGIN domain.example.

  @         IN  SOA  […]

  hello     IN  TXT  "lowercase"
  HELLO     IN  TXT  "uppercase"

  sub       IN  NS   ns1.somewhere-else.example.
  SUB       IN  NS   ns2.somewhere-else.example.

With the default settings, only one of the two TXT records was served
when the server receives a query for hello.domain.example, and only one
of the two name servers of the referral was served for a query of
anything at or below sub.domain.example.

Now, all of the nodes with equivalent case are always retrieved (except
if case-sensitive matching is enabled for a subtree).

Introduce a family of functions in Catalog: fetch_all/2 and /3 gets all
keys and values where the keys are equal or only differ by case, and
fetch_all_values/2 and /3 which only gets the values.

Refactor the case mapping logic into a separate structure and a few
separate companion functions. Doing so will make it possible to
introduce another fetch/3-like function that returns all data with their
original cases.