avx2: [aes-gcm] improve performance for larger buffers

- re-arranged GHASH key table layout
- added label alignment and move 1st data block load to after counter blocks creation
- code size reduction
  - added GHASH last 7 and last 8 functions
  - initial blocks implemented as a function that works across different key sizes
- improvements in GHASH_8_ENCRYPT_8_PARALLEL:
  - do not explicitly load hash keys into register, use memory operand with pclmulqdq
  - pipeline loads of data for GHASH, all key loads are aligned
  - re-worked last encryption round, load, xor and store of plain/cipher text
  - made use of extra T4:T7 registers to load plain/cipher text