fix: Fixed flops estimation for pooling

[frgfm]

Previously, integer output size for adaptive pooling was not supported:

from torchscan import summary
import torch.nn as nn
summary(nn.Sequential(nn.Conv2d(3, 32, 3), nn.AdaptiveAvgPool2d(1)).cuda(), (3, 256, 256))

would produce

~/Documents/torch-scan/torchscan/modules/flops.py in flops_adaptive_avgpool(module, input, output)
    220     # Approximate kernel_size using ratio of spatial shapes between input and output
    221     kernel_size = tuple(i_size // o_size if (i_size % o_size) == 0 else i_size - o_size * (i_size // o_size) + 1
--> 222                         for i_size, o_size in zip(input.shape[2:], module.output_size))
    223 
    224     # for each spatial output element, sum elements in kernel scope and div by kernel size

TypeError: zip argument #2 must support iteration

This PR fixes the issue by casting the output size to a tuple of integer of the correct size.

[codecov]

Codecov Report

Merging #19 into master will increase coverage by 0.04%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #19      +/-   ##
==========================================
+ Coverage   89.10%   89.14%   +0.04%     
==========================================
  Files           8        8              
  Lines         468      470       +2     
==========================================
+ Hits          417      419       +2     
  Misses         51       51              

Impacted Files	Coverage Δ
torchscan/modules/flops.py	99.01% <100.00%> (+0.01%)	⬆️