Add a Linux Shell that uses GTK for rendering.

[robert-ancell]

This is a start on a Linux shell, and only supports rendering of Flutter, no
input is currently provided. The API is expected to change as we flesh out
the shell. The shell doesn't build by default yet, to enable it you need to
set the build_linux_flag gn build argument.

This is the first stage of flutter/flutter#30729

The shell only works on X servers, the EGL code will need to be updated to cover
Wayland.

The Flutter engine currently crashes when the widget is destroyed, but this
seems to be an issue in the engine.

The code is designed to match GTK conventions.

A minimal GTK application that uses this looks like:

int
main (int argc, char **argv)
{
    gtk_init (&argc, &argv);

    GtkWidget *window = gtk_window_new (GTK_WINDOW_TOPLEVEL);
    gtk_widget_show (window);

    g_autoptr(FlDartProject) project = fl_dart_project_new ("./build/flutter_assets", "./linux/flutter/ephemeral/icudtl.dat");
    g_autoptr(FlView) view = fl_view_new (project);

    gtk_widget_show (GTK_WIDGET (view));
    gtk_container_add (GTK_CONTAINER (window), GTK_WIDGET (view));

    gtk_main ();

    return 0;
}

[stuartmorgan-g]

Sorry for the long delay; I was OOO for several days.

Capturing from some offline discussion for anyone looking back at this wondering about the style: because this is intended to be consumed at the GTK level, we have a fairly strong constraint to follow GTK style to avoid it being extremely alien to anyone familiar with GTK conventions. Unfortunately that means we have to violate the Google C++ style guide in a number of ways (function naming, use of macros in API declaration). Given that we have to do that in the API, we're being consistent about those things in implementation details as well to avoid the internals being too much of a hybrid that is hard to understand for someone used to either style.

However, where there are internal-only details that aren't strongly at odds with the rest of the style, we'll follow Google C++ style. Examples:

• Using C++, not pure C, for the implementation files.
• Forbidding C-style casts.
  But we won't, for instance, use Google-style (camel case) function names for static functions in the same files as we're using snake case for the public functions.

[stuartmorgan-g]

As a follow-up PR, it would be good to move this to align with #17210. At the level of this API, we should enforce a folder structure for all the artifacts we expect (both to simplify the API, and to make it more future-proof), and since we're writing the tooling that people will be expected to use to generate that folder they won't even need to know the details.

[robert-ancell]

I've made this change in another branch, but can fold it into this PR if you like.

[stuartmorgan-g]

Filed flutter/flutter#53176 for the work necessary to have this build by default. The earlier that happens the better, as it'll give us bot coverage for the GTK PRs.

[robert-ancell]

I noticed the commit message is completely different - is that expected?

[stuartmorgan-g]

I noticed the commit message is completely different - is that expected?

Ugh, I forgot that the 'waiting' label triggered the auto-commiter. It has terrible commit message handling at the moment (which is the main reason I don't usually use it).

Luckily our commits point back to the PRs, so it's not too hard to get better info. But I do generally put the PR description into the commit message instead of the auto-generated squash message.

[robert-ancell]

This is being reverted, right? Should I make a new PR after that occurs?

[stuartmorgan-g]

No, there was some confusion because an infra failure on the bots happened to show up on the build with this PR. It shouldn't actually be reverted.