(Don't merge until we find the right format for depth)

This PR improves speed of simgetimages by :

• changing pixel format to RGBA8 in PIPCamera.cpp following suggestion Improve image capture speed #1818, and
• only copying RGB in RenderRequest.cpp in these lines https://github.com/Microsoft/AirSim/blob/master/Unreal/Plugins/AirSim/Source/RenderRequest.cpp#L109-L118
  I've made changes from RGBA->RGB in the client side code I could find.
  An added benefit of RGBA -> RGB is that the image responses from the simgetimages() API match the unreal subwindow previews nicely. This has been a complaint by our users in the past, in that the simgetimage() responses tend to be much darker than the preview in the subwindow

Let's talk about RGB first:

• Behind the scenes, Unreal initializes the render target to RGBA16f as can be seen in the engine code here

• The suggestion in Improve image capture speed #1818 works well in that we can initialize it to EPixelFormat::PF_B8G8R8A8.
  There's also an EPixelFormat::PF_R8G8B8A8_UINT, but trying that makes the engine crash with assert failure in that the requested render target format is not supported.
  Interestingly, PF_R8G8B8A8 leads to the same assert failure.
  There's no PF_B8G8R8A8_UINT, unfortunately.
  Side note: these names are misleading in that BGR is actually RGB - https://answers.unrealengine.com/questions/69615/are-epixelformats-pf-b8g8r8a8-and-pf-r8g8b8a8-flip.html
  Why engine crashes though, is something I am not sure about.

• Q. We should try to get int8 images / textures? We can hopefully change the scene capture to get rgba int (not float) and then use PF_R8G8B8A8_UINT? Although I am not sure if this will help coz the GPU would have the images as floats only.

Moving on to depth:

• Now, there should be a similar optimization possible for render targets for depth images. As far as I understand, it should be EPixelFormat::PF_R16F as alluded by this line https://github.com/Microsoft/AirSim/blob/master/Unreal/Plugins/AirSim/Source/RenderRequest.cpp#L127. However doing so, again leads to engine crashing with assert failure in that the requested render target format is not supported .
  Not completely sure why is that++
  My hunch is this happens coz the capture source is being initialized to ESceneCaptureSource::SCS_FinalColorLDR for all image types at this line https://github.com/Microsoft/AirSim/blob/master/Unreal/Plugins/AirSim/Source/PIPCamera.cpp#L66.

• ESceneCaptureSource::SCS_FinalColorLDR doesn't seem efficient for depth images. I tried changing it to ESceneCaptureSource::SCS_SceneDepth, similarly there's a ESceneCaptureSource::SCS_Normal which should be used for the normal image?
  Unfortunately, engine crashes yet again with those two options.

• Anyhow, for now, I have added a placeholder TMap for image_types to pixel_format types. The map is only used for all non depth image types (aka image_type==0 || image_type==5 || image_type==6 || image_type==7).
  for depth images, I still call initautoformat for now.

P.S. Technically, we should use ETextureRenderTargetFormat, which is the set of EPixelFormats exposed to rendertarget.
However, the mapping is fairly intuitive from the variable names as can be seen in the engine code in these lines, so, i am sticking with pixelformat