[FrontEnd] add control flow, basic tensor array support for paddle faster_rcnn model

[luo-cheng2021]

Details:

• Add following op:
  • control flow
    • while
    • conditional_block
    • select_input(dependent by control flow)
  • tensor array
    • lod_array_length
    • tensor_array_to_tensor
    • write_to_array
• transformations to support control flow and tensor array inside paddle frontend
  • TransformIf
  • TransformTensorArray
  • TransformWhile
• add dynamic output to Loop to support the snippet:

x = np.array([1,2])
s = [x,]
i = 0
while i < 10:
    if i > 3:
       s.append(x)
    i += 1

• add the ability to turn off 'RemoveConcatZeroDimInput'
• adjust the paddle frontend to support subblock

Tickets:

• 69931
• 71854
• 69930

[luo-cheng2021]

@ceciliapeng2011 @liubo-intel @meiyang-intel @ilyachur could you please help to have a look, thanks

[luweizhou2016]

Okay for me. Thx!

[yuxu42]

@mbencer Could you please take a look at the comment from @luo-cheng2021 and see any further concern? Thanks!

[luo-cheng2021]

@slyalin @itikhono could you please take a look again, thanks.

[ceciliapeng2011]

@slyalin do you have more comments about it? thanks.

[jane-intel]

@luo-cheng2021 could you please resolve conflicts?

[luo-cheng2021]

@luo-cheng2021 could you please resolve conflicts?

Done.