Fix ImDrawList::AddImageRounded

[Soreepeong]

The function makes an assumption that there exists 1 font atlas texture, so that ImDrawList._Data->TexIdCommon ==
ImDrawList._CmdHeader.TextureId. Since we support multiple font atlas textures, that assumption is no longer true and
ImDrawList::AddConvexPolyFilled will create a new draw command as needed, giving ImGui::ShadeVertsLinearUV a clean draw command to work with.

This workaround forcefully sets the font atlas texture to be the texture the user is trying to draw for the duration of drawing polygons and shading those vertices again, so that no draw command change happens. Once the operation is done, font atlas texture is reverted back to what it was.

This fix is done without thread safety concerns, but an ImDrawList should not be touched from multiple threads at a single time, so this is fine.

[Soreepeong]

https://github.com/goatcorp/gc-imgui/blob/e4920dc10cbf32c0cd9b4a48a4a71c8814242460/imgui_draw.cpp#L1669-L1693

ImDrawList::PathFillConvex in turn calls ImDrawList::AddConvexPolyFilled, which will push and pop the common texture ID as needed during its scope, which will often end up adding an empty draw command after it's done.
https://github.com/goatcorp/gc-imgui/blob/e4920dc10cbf32c0cd9b4a48a4a71c8814242460/imgui_draw.cpp#L993-L1001

ImGui::ShadeVertsLinearUV works on the most recent draw command.
https://github.com/goatcorp/gc-imgui/blob/e4920dc10cbf32c0cd9b4a48a4a71c8814242460/imgui_draw.cpp#L1909-L1931

[KazWolfe]

At what point are we just going to hardfork ImGui?

[Soreepeong]

We already did; we're just trying to not rebuild cimgui.dll for the time being. Proper fix should go in with ImGui version update