Add guppy backend

[dxe4]

related to my comment here #236 (comment)
with implementation of guppy3 we can get the right values.

note this implementation does self.prevlines[-1] == self.prev_lineno: for guppy backend only, because repeating [i for i in range(0, 10000000)] makes the guppy backend extremely slow. although this may not be ideal or acceptable i thought it's worth sharing the code.

@profile
def myfn_py():
    ones = [1] * NBR_ITEMS
    twos = [i + 1 for i in ones]
    return twos


@profile
def myfn_np():
    ones_np = np.ones(NBR_ITEMS)
    twos_np = ones_np+1
    return twos_np


myfn_np()
myfn_py()


python -m memory_profiler --backend guppy prof.py 


Line #    Mem usage    Increment   Line Contents
================================================
    11   14.191 MiB   14.191 MiB   @profile
    12                             def myfn_py():
    13   18.006 MiB    3.815 MiB       ones = [1] * NBR_ITEMS
    14   22.098 MiB    4.091 MiB       twos = [i + 1 for i in ones]
    15                                 return twos


Filename: prof.py

Line #    Mem usage    Increment   Line Contents
================================================
    18   14.189 MiB   14.189 MiB   @profile
    19                             def myfn_np():
    20   18.005 MiB    3.817 MiB       ones_np = np.ones(NBR_ITEMS)
    21   21.820 MiB    3.815 MiB       twos_np = ones_np+1
    22   21.820 MiB    0.000 MiB       return twos_np

[dxe4]

this can cause issues if you want to measure the memory of a side effect eg

[foo() for i in range(0, 500)]
def foo():
     make_big_array = [i for i in random.randint(0, int(1e9))]

i haven't tested it.
although it will give correct results for this case #236 (comment)