This is WIP to move IPython's backend mapping into Matplotlib and
extends it to allow backends to register themselves via [entry
points](https://setuptools.pypa.io/en/latest/userguide/entry_point.html#entry-points-for-plugins).
It is a coordinated effort across Matplotlib, IPython, matplotlib-inline
and ipympl. Closes #14311. Most of the work is in Matplotlib
(matplotlib/matplotlib#27948) but I will repeat the relevant points
here.

When using a Matplotlib magic command of the form `%matplotlib
something` the identification of the Matplotlib backend and GUI loop are
now performed in Matplotlib not IPython. This supports:

1. Matplotlib backends that IPython already supports such as `qtagg` and
`tkagg`.
2. Other built-in Matplotlib backends such as `qtcairo`.
3. Backends that use `module://something` syntax such as
`module://mplcairo.qt`.
4. Backends that self-register using entry points, currently `inline`
and `widget`/`ipympl`.

Implementation details:

1. The magic command is now explicitly `%matplotlib gui_or_backend`
rather than `%matplotlib gui`. This is already effectively the case as
e.g. `%matplotlib ipympl` is really a backend not GUI name. Within
Matplotlib the `gui_or_backend` is checked first to see if it is a GUI
name and then falls back to checking if it is a backend name.
2. If you select the `inline` backend the corresponding GUI is now
`None` not `inline`. All backends which do not have a GUI loop return
`None`, otherwise we have to keep explicit checks within IPython for
particular backends.
3. `backends` and `backend2gui` are now deprecated but are still
exposed, with a deprecation warning, if required. If using Matplotlib,
ipympl, etc releases that include the corresponding changes to this PR
then they are not needed as Matplotlib deals with it all. But for
backward compatibility they must still be available for a while along
with the remainder of the existing backend-handling code.
4. I haven't yet updated the documentation but we need to keep
information about valid GUI frameworks and I propose that we should
remove all lists of valid Matplotlib backends, replacing them with
instructions on how to obtain the current list (pointing to the
Matplotlib docs and using `%matplotlib --list`). If we keep any lists
then they will inevitably become out of date. This extends to the
`backend_keys` in IPython/core/shellapp.py.

Because the four related projects are loosely coupled without direct
dependencies on each other (except for `ipython` and
`matplotlib-inline`), backward compatibility requires all possible
combinations of projects before and after the new functionality (I will
call these "old" and "new" from now on) to continue to work. I have
tested these all locally, and the CI of this PR will test new IPython
against old Matplotlib for example, but I need to add one or more new
temporary CI runs to test new IPython against new Matplotlib etc. The
identification of new versus old depends on version checks on the other
libraries, so here is a table that I will update showing the current
status of progress in the 4 projects:

| Project | Relevant PRs | Possible release version |
| --- | --- | --- |
| matplotlib-inline | ipython/matplotlib-inline#34,
ipython/matplotlib-inline#35 | 0.1.7 |
| ipympl | matplotlib/ipympl#549 | 0.9.4 |
| Matplotlib | matplotlib/matplotlib#27948 | 3.9.1 |
| IPython | #14371 (this) | 8.24.0 |

The two widget projects can be released soon, once we are happy with the
entry point approach. The other two projects' PRs will have to be
synchronised as each includes version checks on each other.

To do

- [ ] Add CI runs against the new PR branches of the other projects.
- [ ] Add comments for conditions required for backward-compatibility
code blocks to be removed.
- [ ] Update documentation, including removal of lists of valid
backends.
- [ ] Update version checks before merging.

Fixes #14401 which has been a bug in the 8.22.x and 8.23.x releases.

When I removed the multiple initialisation of Matplotlib backends in
#14330 it broke use of the following:
```bash
ipython --matplotlib
ipython --matplotlib=auto
ipython --pylab
ipython --pylab=auto
```
by failing to display Matplotlib plot. If you specify a particular GUI
event loop such as using
```bash
ipython --pylab=qt
```
then it was and is fine. So for anyone finding this, the workaround
until the next release is to specify a GUI loop rather than relying on
the auto selection.

I didn't notice this as I've been concentrating on moving the Matplotlib
backend logic from IPython to Matplotlib, and with those changes
(matplotlib/matplotlib#27948) the above use cases all work OK.

The fix is to reintroduce the early import of `matplotlib-inline` but
only if both the gui loop is not specified and the Matplotlib version is
before the movement of the backend logic across to it.

There are no explicit tests for this. In the future I will try to think
of some tests for some of this IPython-Matplotlib functionality that
don't involve installing complicated backend dependencies or adding
image comparison tests.

…lib and support entry points

…948-on-v3.9.x

Backport PR #27948 on branch v3.9.x (Move IPython backend mapping to Matplotlib and support entry points)

This is a correction to the code that moves backend handling to
Matplotlib, which came to light during the review of
matplotlib/matplotlib#27948. The GUI framework of macOSX is called `osx`
in IPython but `macosx` in Matplotlib.

It would be possible to allow passing a GUI framework of `osx` to
Matplotlib and internally converting it to `macosx`, but the reverse
direction is problematic as we would like the answer to be `osx` if we
are using IPython and `macosx` if using pure Matplotlib. Therefore the
simplest solution is to do the translation in IPython. It is not ideal
as we want to minimise such Matplotlib-related implementation details in
IPython, but the changes here are small and simple and hence I think
they are acceptable.

There are two new private functions `_convert_gui_to_matplotlib` and
`_convert_gui_from_matplotlib` to do the required checking and
conversion, and these are called whenever a GUI framework name is passed
to or from Matplotlib. There are only 3 places in the code where this is
currently required.

Inevitably this comes to light just after the release of IPython 8.24.0!
But it is not a problem for end users until the next Matplotlib release
which contains matplotlib/matplotlib#27948. If that occurs really
quickly (sometime in May?) perhaps we could release an IPython 8.24.1
just beforehand, otherwise the usual planned release at the end of next
month would be fine.