work around missing reshape() callbacks from LWJGL v3 (issue #1793)

[stephengold]

Based on the observations documented in #1793, the key issue is that transitions between windowed mode and fullscreen mode in LWJGL v3 don't always generate reshape() callbacks. This spoils GUI layouts that need updating each time the dimensions of the graphics context change. The user sees a layout generated for a different-sized framebuffer.

(There's also an issue with the initial reshape() callback, which occurs on Windows, but not Linux, when the initial context is created. However, that's an easy issue for apps to work around.)

The key issue manifests differently on Linux than on Windows:

• On Linux LWJGL v3, windowed-to-fullscreen causes a callback, but not fullscreen-to-windowed.
• On Windows LWJGL v3, neither transition causes a callback.

This PR proposes that jme3-lwjgl3's LwjglWindow should explicitly invoke reshape() on both windowed-to-fullscreen and fullscreen-to-windowed transitions. This results in some redundant callbacks (multiple callbacks for a single createContext()). However, I expect most reshape() callbacks perform idempotent operations. Also, filtering out redundant callbacks is much easier that compensating for missing callbacks.

[stephengold]

Lying in bed this morning, I came up with a simpler, more elegant fix.

[stephengold]

Here's a summary:

• deal with all size/scale changes in a single method, now named updateSizes()
• keep a private record of the framebuffer size, so updateSizes() won't be confused by unrelated changes to AppSettings

[stephengold]

Unless there's more to discuss, I propose to self-integrate this PR in about 24 hours.

[stephengold]

In order to backport this into "v3.5" branch, we'd also need to backport #1750. The combination of the 2 PRs seems risky relative to the benefits. Better wait for 3.6 or 4.0

[Ali-RS]

Looks like this breaks TestAWTPanels, please see #1945