For some BGRA-backed formats (RGBA4, R5G6B5, RGB5A1), our ReadPixels
implementation wasn't aware the BGRA format didn't exactly match the
RGBA format. For these it would do the 'fast path' memcpy method, when
it should stop and do the slow pixel-by-pixel packing method.

Fixes conformance2/reading/read-pixels-from-fbo-test.html.

Reland: fix empty format info that was causing us to only see the
first pixel in the FBO in a ReadPixels call. Also fix bugs in the
unorm 16-bit format readback code, and add ASSERTs to catch bugs in
subsequent new formats.

BUG=angleproject:1407
BUG=chromium:616176

Change-Id: I9fd55b9e1dd6a306eb4db195d775c02a1eb1f93f
Reviewed-on: https://chromium-review.googlesource.com/357132Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>