Fix bin comprehensions when option is used

[sabiwara]

Fix #15460

Addresses both the regression on integer and the original inconsistency issue for variables/operations.

iex> k = 8;  bin = <<0, 1, 9, 2, 0, 3>>
iex> for(<<0::8, x::8 <- bin>>, uniq: true, do: x)
[1, 3]
iex> for(<<0::8, x::size(^k * 1) <- bin>>, uniq: true, do: x)
[1, 3]
iex> for(<<0::8, x::size(^k) <- bin>>, uniq: true, do: x)
[1, 3]

I don't know if there is a better way to avoid having to drill down the variables.
This is the only way I found to distinguish between a variable that is available from before the comprehension or a var from the generators.

[sabiwara]

OK, I think this is now ready to ship.

[josevalim]

I'd perhaps add a test for nested generators and we are good to go!!! Awesome job and please backport!!

[sabiwara]

@josevalim is it what you had in mind by nested generators? dd490a9

[josevalim]

Yes!!!!

[sabiwara]

Backported ✅