faster xor_from_ptr using fixed 16/256 return address

[tomhea]

remove the double call to ptr_wflip_second_word (~3w@) by jumping to a fixed-address 16/256 that also holds the read-from-ptr-handling.

There are 3 ways of making the (*ptr).jump = 16/256:

• Replacing the ptr_wflip_second_word with a slimmer macro: flip-ptr-2nd-word-4/8th-bit (total: ~w@).
• Flipping a big chunk of memory by rep(30000, i) fj mem+i*dw+dbit+4/8, $, or define that memory section in the first place with the 4/8th bit on.
• Creating .fjm v4 that supports placing a fixed fj-op multiple times (can also allow other different default-op than 0) on the way. It will also require fj-parser new command: reserve_op num_of_ops, fixed-flip-address, fixed-jump-address (e.g. reserve_op 30000, 0, 256 for initializing brainfuck memory)

My preferable choice is (*ptr).jump = 256 (byte-read, which also supports hex/bit reads) + creating .fjm v4 + reserve_op command. Also, implement a xor_byte_to_ptr + xor_ptr_to_byte.

[tomhea]

I implemented (*ptr).jump = 256 (byte-read, which also supports hex/bit reads), and I didn't create a .fjm v4 as I just flipped the 256 bit.

I also implemented read_byte / write_byte, and same for vectors

I also transformed the push n / pop n to use the read/write_byte n.

merged in #250.