Using .update has a memory leak

[pashashocky]

Description

Using .update method of a bounter counter with anothe counter has a memory leak.

Steps/Code/Corpus to Reproduce

printing would be useful, but here's the idea

def mem_leak():
    get_mem()
    c1 = bounter.bounter()
    c2 = bounter.bounter()
    for i in range(10**7):
        c1[random.randint(0,10)] += 1
        c2[random.randint(0,10)] += 1
    get_mem()
    c1.update(c2)
    get_mem()

def get_mem():
    print(psutils.Process(os.getpid()).memory_full_info().rss / 2 ** 20)  # in mb

get_mem()
mem_leak()
get_mem()
mem_leak()
get_mem()

Expected Results

Expect counters c1 and c2 to completely be destroyed from memory after mem_leak function exits.

Actual Results

Some memory remains used even after mem_leak function exits.

Versions

can't run versions, but:
Linux 4.14 amzn linux x86
Python 3.6.6
Numpy 1.16.4
pandas 0.24.2
bounter 1.1.0

@mpenkov