In CoreGraphicsPixelData::createImage, image data was copied from a
BitmapData created from the Image passed into the function.

The BitmapData instance didn't keep track of the size of the buffer it
pointed to, so the buffer size was computed by multiplying the
BitmapData height by its line stride. However, if the BitmapData pointed
to a subsection of an image, the `data` pointer might be offset from
the allocated region, and `data + lineStride * height` would point past
the end of the allocated region. Trying to read/copy this range would
cause a heap buffer overflow at the end of the range.

This change adjusts BitmapData so that it keeps track of the size of the
allocated region. Taking a subsection of an image should subtract the
data pointer offset from the size of the allocated region.

Narrator on Windows seems to call ExpandToEnclosingUnit to find the
character immediately following the current cursor position. When
expanding a degenerate range with position `input` (used to represent
the cursor position) to a single-character range, the returned range
should start at `input` and end at `input+1`. Previously,
findTextBoundary() would always return the position before
currentPosition when searching backwards by character, so the result of
ExpandToEnclosingUnit would be off-by-one when expanding to the closest
character.

The bug was triggered on Monterey where a pressure of 1 is reported
while a mouse button is being held down. This caused an extra drag
event being triggered between mouse down and up events, even if no
movement occurred.

This change is intended to resolve an issue where iOS apps failed
automated review due to use of non-public selectors. Explicitly adopting
the relevant protocols allowed an app to pass automated review during
testing.

This issue affected windows with JUCE titlebars. When dragging the
window back and forth between two displays with different DPI settings,
the window could sometimes end up with the wrong size.

It looks like setting new window bounds inside the WM_WINDOWPOSCHANGED
message was occasionally causing the system to lose track of the correct
scale for the window.

As a byproduct of this change, JUCE should no longer receive reentrant
calls to handleDPIChanging.

With this patch applied, the DemoRunner should build under MinGW, and be
(nearly) feature-complete compared to the MSVC build.

Specifically, when building with MinGW:
- Adds support for accessibility
- Fixes build issues in the juce_video module
- Fixes a link issue in the VST3 wrapper when VST3_CAN_REPLACE_VST2 is
  defined
- Adds support for the new-style native FileChooser
- Tidies up some other low-severity warnings

Known issues:
- Direct2D rendering is still not supported when building with MinGW due
  to ABI compatibilities.

Effects of this change for windows with native titlebars:
- Resizable windows without maximise buttons will now only allow the
  'zoom' behaviour from the rightmost titlebar button (i.e. this button
  will not allow entering full-screen mode)
- Non-resizable windows will grey-out the rightmost titlebar button,
  whether or not the maximise button is enabled

This works around a rare issue in the DialogsDemo where the AlertWindow
would fail to display because no peer had focus.

When FileChooser instances were created, launched, and hidden all inside
the same event callback on macOS 12.0.1, the chooser dialog sometimes
remained open. This could cause problems including crashes, as closing
the dialog would attempt to call a completion handler block referencing
an already-deleted FileChooser::Native instance.

Opening the chooser panel later on the message thread seems to resolve
the issue.