fix: Upcast small integer dtypes for rolling sum operations

[lukemanley]

fixes #21318

Updates Expr.rolling_sum and Expr.rolling_sum_by to upcast small integer dtypes to Int64 to be consistent with other sum operations such as Expr.sum and Expr.cum_sum.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.95%. Comparing base (3719977) to head (bfdbf31).
Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #21397      +/-   ##
==========================================
+ Coverage   79.89%   79.95%   +0.06%     
==========================================
  Files        1597     1597              
  Lines      228951   228968      +17     
  Branches     2618     2618              
==========================================
+ Hits       182915   183081     +166     
+ Misses      45437    45288     -149     
  Partials      599      599              

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

[wence-]

It be nice if this cast could be promoted into the logical plan type coercion phase so that we don't have to mimic this behaviour in the gpu engine too.