Specialized fpc_varset_create_element and fpc_varset_(un)set for i386 and x86_64.

These fpc_varset_(un)set are faster than current Include / Exclude and after the recent change the benchmark has to make an effort to not get Include for set += [a] :D But this is a shortcoming of Include / Exclude that hopefully will be addressed someday. (No more after 12176d36.)

Specialized versions should be useful regardless. (Especially the day compiler learns to generate unset...)

Benchmark (of System functions): Set1Benchmark.pas.

My results:

x64                                            before          after
StringToCharSet (varset_set):                563 ns/call    479 ns/call
StringToCharSet (include):                   192 ns/call
StringToCharSets (varset_create_element):    852 ns/call    469 ns/call

i386                                           before          after
StringToCharSet (varset_set):                800 ns/call    482 ns/call
StringToCharSet (include):                   398 ns/call
StringToCharSets (varset_create_element):    0.9 us/call    781 ns/call