Fix description and formula in pooling operations

[zolkis]

Originally posted by @shiyi9801 in #446 (comment)

What is filter size in pooling operations, do you mean window_size?
Also this equation seems wrong. Referring to conv2d operation, I guess the correct equation looks like this?
output_size = ceil(1 + (input_size - (window_size - 1) * dilation - 1 + beginning_padding + ending_padding) / stride)
So as floor.
See the Returns section in https://webmachinelearning.github.io/webnn/#api-mlgraphbuilder-pool2d

[zolkis]

This seems not specific to #446, it's a separate problem.
We need to check the formula in the Returns section.
Then we can update the algorithm. Right now the algorithm uses generic formulation, leaving to the implementation how to calculate the output size.

[anssiko]

This part of the Returns section:

More specifically, if the options.roundingType is "floor", the spatial dimensions of the output tensor can be calculated as follow:

output size = floor(1 + (input size - filter size + beginning padding + ending padding) / stride)

or if options.roundingType is "ceil":

output size = ceil(1 + (input size - filter size + beginning padding + ending padding) / stride)

... reads like informative text. It should be marked as such and the two formulas corrected.

[zolkis]

Fixed by #478