The following was observed for a VST3 plugin hosted in Live 11.1 with
auto-scaling disabled:

- It never calls setContentScaleFactor on the plugin's UI, so the
  wrapper has to check the current display on a timer and update the
  current scale factor when necessary.

- It calls canResize on the plugin view after opening it, but doesn't
  seem to respect the result of this call. According to the VST3
  documentation, a host is supposed to only call checkSizeConstraint
  during a live resize operation (which should only happen if the plugin
  reports it can resize), but Live calls this function every time the
  user drags the editor. It also passes the result of this function to
  onSize, whether or not checkSizeConstraints reported success.

- When dragging an editor between displays, Live will continue to call
  checkSizeConstraint and onSize with the editor’s old size in physical
  pixels. In some cases, JUCE's "scale factor check" timer callback
  fires, resizes the view to the correct size, and then Live
  asynchronously calls onSize again with the editor's old size in
  physical pixels, resulting in the editor being set to the wrong
  logical size.

This patch ensures that checkSizeConstraint always returns the current
size of a nonResizable editor. This means that the logical size of the
editor should not change when the result of checkSizeContraint is used
to resize the window.

This partially reverts d9c25ec17d.

When calling NullCheckedInvocation::invoke with a capture-less lambda,
GCC 9.3 determines that the expression
    std::declval<TheLambda>() != nullptr
is well-formed, and uses the version of invoke containing a nullptr
check. However, the compiler is also able to determine that this
expression can never be false, and emits a warning.

MIDI only plugins are no longer provided valid audio buffers
Graph nodes are only executed when they have active connections

Fix improper use of non-atomic pointer validity to report state of background search thread state

Following commit 093dbc7df1 pings of a
child process connection need to initialised explicitly. This commit
adds a missing initialisation.

Previously it was possible to have a dangling resource map. In that case
opening an AU from a file that did not have a resource map would lead to
using the previously unclosed map and attempting to load the previous
plugin again.

fc378aaf9a introduced a regression where
plugins with no audio channels (such as MIDI FX plugins) would receive
an audio buffer with a length-in-samples of '0', rather than the actual
block length.

This change also allows the plugin to receive discrete layouts in hosts
that support them, while also maintaining support for
AudioChannelLayoutTags in Logic.

These were mistakenly removed in afe5199848

With clang 13.0.0, and Apple clang version 13.1.6 (clang-1316.0.21.2),
the following code fails to compile with `-std=c++20 -O3 -msse3`:

    #include <immintrin.h>

    auto test (__m128 a)
    {
        return _mm_hadd_ps (_mm_hadd_ps (a, a), a);
    }

This partially reverts f43784dc: "macOS:  Propagate focus loss message when resigning key window status and only grab focus if window can become the key window"

The change introduced in 92f350e617 led to
a regression in the MoveEndpointByUnit function. In this case, a
backward movement by a single character *should* move the endpoint as
long as it remains within the text range.

The issue addressed by the faulty commit is better fixed by
special-casing the 'character' unit case in the ExpandToEnclosingUnit
function.