Update t_vertex_generic.c: Don't do inefficient sharded copies that can be...

mesa/tnl: use memcpy for more efficient copies

Here we replace inefficient copies that can be done in a single instruction.

Diff for Function insert_4f_4

00536700:    8b 02    movl (%rdx), %eax    ~~~> 00536700:    0f 10 02 movups (%rdx), %xmm0
00536702:    89 06    movl %eax, (%rsi)    ~~~> 00536703:    0f 11 06 movups %xmm0, (%rsi)
00536704:    8b 42 04 movl 0x4(%rdx), %eax ~~~>                                           
00536707:    89 46 04 movl %eax, 0x4(%rsi) ~~~>                                           
0053670a:    8b 42 08 movl 0x8(%rdx), %eax ~~~>                                           
0053670d:    89 46 08 movl %eax, 0x8(%rsi) ~~~>                                           
00536710:    8b 42 0c movl 0xc(%rdx), %eax ~~~>                                           
00536713:    89 46 0c movl %eax, 0xc(%rsi) ~~~>                                           
00536716:    c3       retq                 ===> 00536706:    c3       retq