fix: zip_with also broadcast mask

[reswqa]

Per discuss in discord, and It seems that all test can be passed.

[ritchie46]

Well... That was easy! 😄

[ritchie46]

@reswqa shall we take this opportunity to write it via ternary_elementwise. Those will be better optimized.

[reswqa]

shall we take this opportunity to write it via ternary_elementwise.

@ritchie46 As for the impl_ternary_broadcast branch, I guess we only need apply and binary_elementwise.

When I tried to do this, I ran into lifecycle issues with utf8/binary types.

Take the following as example:

(l_len, 1, mask_len) if l_len == mask_len => {
            let right = $other.get(0);
            let val:ChunkedArray<$ty> = binary_elementwise($mask, $self, |mask:Option<bool>, left|{
                ternary_apply(mask.unwrap_or(false), left, right)
             }).with_name($self.name());
            Ok(val)
},

error: implementation of `FnOnce` is not general enough
   --> crates/polars-core/src/chunked_array/ops/zip.rs:76:41
    |
76  |               let val:ChunkedArray<$ty> = binary_elementwise($mask, $self, |mask:Option<bool>, left|{
    |  _________________________________________^
77  | |                 ternary_apply(mask.unwrap_or(false), left, right)
78  | |              }).with_name($self.name());
    | |_______________^ implementation of `FnOnce` is not general enough
...
174 | /             impl_ternary_broadcast!(
175 | |                 self,
176 | |                 self.len(),
177 | |                 other.len(),
...   |
183 | |             )
    | |_____________- in this macro invocation
    |
    = note: closure with signature `fn(std::option::Option<bool>, std::option::Option<&'2 [u8]>) -> std::option::Option<&[u8]>` must implement `FnOnce<(std::option::Option<bool>, std::option::Option<&'1 [u8]>)>`, for any lifetime `'1`...
    = note: ...but it actually implements `FnOnce<(std::option::Option<bool>, std::option::Option<&'2 [u8]>)>`, for some specific lifetime `'2`

[cmdlineluser]

I ran into lifecycle issues

I think that's the same issue I ran into.

@orlp showed me how to fix: #12071 (comment)

[ritchie46]

@reswqa I think we can follow up with the improvement on a second PR. This PR is the fix. Would love to see if we can make it better with proper column traversal. :)