backends/drm: fix multi gpu with mixed explicit and implicit modifiers
Commit 1 fixes EglGbmLayerSurface::doesSurfaceFit to more explicitly handle multi gpu modes, so that it never happens that the surface gets recreated more often than necessary
Commit 2 fixes the logic for when modifiers are used and the linear modifier is forced - before it forced a linear modifier on the presenting GPU with EGL copy, which is nonsense, and did not force it for the source GPU when needed