GCM VAES/VPCLMULQDQ performance improvements for large buffers (issue #35)

- GCM_BIG_DATA compile flag added
  - disabled by default, number of ghash keys is 48 (key structure 1,152 bytes)
  - when ebaled, number of ghash keys is 128 (key sturcture 2,432 bytes)
- precomputing 128 or 48 GHASH keys (the structure is much bigger now)
  - When GCM_BIG_DATA is on and data >= 2,048 bytes reduction is done every 128 blocks
  - for data >= 768 bytes reduction is done every 48 blocks
  - for other cases reduction is done every 8 blocks
- added new macro handling 16 blocks of AES and GHASH in parallel
  - pipeline depth is 32 blocks
  - very large and large buffers leverage the macro
- initial N x 16 blocks macro implemented
  - pipelines cipher with GHASH
  - first runs cipher only as defined by depth of the pipeline
  - then runs stitched cipher and GHASH for the remaining number of blocks
- parallel cipher and ghash N x 16 blocks implemented
  - cipher and ghash always stitched
  - reduction done as defined by maximum number of blocks
  - depth of the pipeline maintained as in initial N x 16 macro
    - cipher is ahead of ghash by 32 blocks
- stack frame created to keep up to 128 blocks of cipher text
  - stack frame loads/stores are aligned
- gcm key data structure definition made more generic
- VX512STR and VX512LDR macros changed from vmovdqu64 to vmovdqu8
  in order to work correctly with masked operations

Change-Id: Idd83b911c9257bbd221c66ddd9297a4f2ae120c2