Miss-hit

* Add another method of tracing Simulink models via [block
properties](https://www.mathworks.com/help/simulink/ug/block-properties-dialog-box.html).

* The top-level system of a Simulink Library (but not Model) no longer
appears in the tracing output, as there is no way to access or
manipulate this in the Simulink user interface.

* Add new option `--untagged-blocks-inherit-tags` for `mh_trace` which
cases any untagged system to inherit any tags from its parent. Once
new tags are encountered this breaks the inheritence chain. Note
that embedded MATLAB blocks are unaffected by this option.

* Fix issue where a Simulink model with self-contained links would not
parse the referenced systems. Only the top-level system would be
parsed and references were never followed.

* Fix issue where a Simulink model with referenced stateflow would not
parse the referenced machines and charts.

* Fix issue in `mh_trace` where sometimes LOBSTER tags could be
generated which contained spaces.

* Support multi-line and HTML annotations in Simulink for `mh_trace`
annotations.

* Rework `mh_trace` and its output format. It now generates LOBSTER
traces. See https://github.com/bmw-software-engineering/lobster for
more information, including a description of the interchange
format. The old output format is no longer supported now that there
is a good standard.

* Add support for Simulink tracing in `mh_trace`. You can add
annotations to any block starting with the text `lobster-trace:`
followed by a list of requirement tags. For example `lobster-trace:
foo.my_req`.

* Remove `mh_trace` commandline flag `--json` and replace it with
`--out-imp` and `--out-act`. The default filename produced is now
`mh_imp_trace.lobster` and `mh_act_trace.lobster`.

* Remove `mh_trace` commandline flag `--by-tag`. You can use a tool
like LOBSTER to recover this information.

* Add `mh_trace` commandline flag `--only-tagged-blocks`. This filters
out all Simulink blocks that do not contain at least one tag.

* Fix a bug in `mh_trace` where precisely duplicated package +
function names result in only one tracable item. Now there are two,
and a tool like LOBSTER will complain.

* Fixed minor issue where `mh_trace` was not made available as a
command-line tool when installing the PyPI package `miss_hit`.

* You can now specify more precise Octave and MATLAB versions. This
change is massive, and likely to have subtle bugs. You can now write
`octave: "4.4"` or `matlab: "2020b"` in your config files; or specify
`--octave=4.4` or `--matlab=2020b` from the command-line. There is
also a special `latest` version for both Octave and MATLAB, which is
an alias for the latest supported version.

This is also the first change that introduces backwards
incompatibility, specifically:
* The command-line option `--octave` no longer works. You need to
specify `--octave=latest` to get the same behaviour.
* The config setting `octave: true` still works, but is
deprecated. It means the same thing as `octave: "latest"`.
* The config setting `octave: false` doesn't make sense anymore (and
never really did), so it now raises an error.

As always note that for MATLAB, support should be fairly good and
accurate. For Octave many things are missing (such as the `end_X`
set of keywords). I do plan to improve the situation, but please
create tickets for things you need sooner.

* Several tools that generate messages (`mh_style`, `mh_metric`, and
`mh_lint`) will now add the originating check id in the message. For
example:


In test.m, line 4
| false = 0.01; % bad
| ^^^^^ check (medium): redefining this builtin is very naughty [builtin_shadow]


This should make it much easier to disable rules if you don't like
to read the manual.

* MISS_HIT now recognises and processes Octave test `.tst` files
(along with `.m` and `.slx` files). The Octave test annotation
language (comments starting with `%!`) is ignored by MH Style for
now. Thank you to Alois Spitzbart for the idea.

* Relaxed docstring recognition: now you can have blank lines (without
the comment indicator) in your docstring.