Handle `is_first` and `is_last` constraint in copy_circuit

[KimiWu123]

Describe the bug

https://github.com/privacy-scaling-explorations/zkevm-circuits/pull/1419#discussion_r1201689532, Han found that there is no constraint to enforce is_first to be 1 and after discussion, there is no constraint for is_last either. More investigating the constraints in copy_circuit, we need to take care the row rotations more carefully. Take value_acc_rlc as example, we take a value from Rotation(2), but we didn't ensure the current row is not the last copy step (which means the next two row could be a next copy event or this copy step is the end of this table).

Conclusion

1. Enforce is_first and is_last to 1 at first and last row properly.
2. Constraints "bytes_left == bytes_left_next + 1 for non-last step" and "value_acc_rlc(2) == value_acc_rlc(0) * r + value(2)" access values from Rotation(2)

---

Just draw a simple example to visualize the relationship of each row in copy_circuit

*Assuming q_enable == true for every row, and ignore is_padand is_code for simplicity.

q_step	is_last	value	value_acc_rlc	copy_table.is_first	copy_table.bytes_left	copy_table.tag/id/src_addr_end
1	0	$byte[0]	$byte[0]	1	n	TxCalldata/$txID/$cdLength
0	0	$byte[0]	$byte[0]	0	-	Memory/$callID/-
...
1	0	$byte[i]	$value_acc_rlc[i]	0	n - i	TxCalldata/$txID/$cdLength
0	0	$byte[i]	$value_acc_rlc[i]	0	-	Memory/$callID/-
...
1	0	$byte[n-1]	$value_acc_rlc[n-1]	0	1	TxCalldata/$txID/$cdLength
0	1	$byte[n-1]	$value_acc_rlc[n-1]	0	-	Memory/$callID/-

*$value_acc_rlc[i] = $value_acc_rlc[i-1] + $byte[i] * r

[ed255]

On a side note, I think it would be interesting to develop a framework to write and evaluate state machine-like circuits so that we spot these underconstraints faster. In this case we could see that there are 4 possible states: First, Middle, Last, FirstAndLast. Each state has some associated expression:

• First: is_first=1 and is_last=0
• Middle: is_first=0 and is_last=0
• Last: is_first=0 and is_last=1
• FirstAndLast: is_first=1 and is_last=1

Then we have transition conditions which constraint what the current or next state is considering signals in the current state; and in this case they are missing. A similar thing happened in the Bytecode Circuit: https://github.com/privacy-scaling-explorations/zkevm-circuits/issues/1332