fix: Fix incorrect lazy schema for explode() in agg()

[nameexhaustion]

Fixes #19562

• Explode during dtype resolution mistakenly indicated that it did not need auto-imploding (see the nested variable in the code)
• Also adjusts the output dtype getting/checking logic for FunctionExpr to adhere to:
  • If the function indicates RETURNS_SCALAR, its output dtype shall always be scalar (i.e. .agg(col(f64).sum()) -> Float64)
  • Otherwise, if we are in a Context::Aggregation, we auto-implode the dtype (i.e. .agg(col(f64).shuffle()) -> list[Float64])
  • Assertion was also adjusted during execution in ApplyExpr
  • To do this properly we also propagate the Context recursively

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.86%. Comparing base (6b0a906) to head (850f15e).
Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #19629   +/-   ##
=======================================
  Coverage   79.86%   79.86%           
=======================================
  Files        1537     1537           
  Lines      211923   211925    +2     
  Branches     2446     2446           
=======================================
+ Hits       169249   169254    +5     
+ Misses      42120    42117    -3     
  Partials      554      554           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ritchie46]

So the fix boils down to correctly passing the proper Context right?

[nameexhaustion]

and also Explode doesn't actually return a scalar so we don't subtract the nested state

[ritchie46]

Interesting. Does that work? It removes a level of nesting?

[nameexhaustion]

I'm not entirely around this nested variable - but it basically does this (you can see above)

if nested >= 1 {
    field.coerce(field.dtype().clone().implode());
}

As I understand it just means we have to set nested to 0 if we don't need auto-implode. I also suspect the exact value of the nested value doesn't matter other than this property, but I'm not sure :/

At the beginning it's initialized to 1 if we are Context::Aggregation and 0 otherwise. So if a result doesn't need an auto-implode (e.g. sum() aggregation), we just set it to 0.

[ritchie46]

Now that I think of it, I think it does need to subtract if nested >= 2.