Add BC7 support. It's incredibly slow - ~60 seconds to compress a 512x512 image, on a Core i7 - but it works.

- Added AVPCL compressor to projects and got it building with VC9 and VC10.
- Removed unused command line interface & file read/write code from AVPCL.
- Convert AVPCL to use NV vector math lib, asserts, etc.
- Convert AVPCL to use double instead of float.
- Added 4x4 symmetric eigensolver, for AVPCL; it's based on the existing 3x3 one, but I had to rewrite the Householder reduction stage.  As with ZOH, using the eigensolver (instead of SVD) gives a ~25% speedup without significantly affecting RMSE.
- Encapsulate ZOH and AVPCL stuff into their own namespaces to keep everything separate.
- Added some missing vector operators to the nvmath lib.

project/vc10/nvcompress/nvcompress.vcxproj

@@ -355,6 +355,9 @@
     <ProjectReference Include="..\bc6h\bc6h.vcxproj">
       <Project>{c33787e3-5564-4834-9fe3-a9020455a669}</Project>
     </ProjectReference>
+    <ProjectReference Include="..\bc7\bc7.vcxproj">
+      <Project>{f974f34b-af02-4c88-8e1e-85475094ea78}</Project>
+    </ProjectReference>
     <ProjectReference Include="..\nvcore\nvcore.vcxproj">
       <Project>{f143d180-d4c4-4037-b3de-be89a21c8d1d}</Project>
       <ReferenceOutputAssembly>false</ReferenceOutputAssembly>