Update projection column by correct index in expression pushdown optimizers - #25431

Thanks for the PR! Two (codex-assisted) comments:

1. Could we let the projection derive its types from its expressions? The index correction fixes the crash, but leaves four places manually updating projection->types in type_pushdown.cpp and scalar_fn_pushdown.cpp.

   The replacement visitors already update the expressions. I would remove those four writes and resolve types once at the end of PushdownOptimize in type_pushdown.hpp:

   if (any_pushed) {
   	Replace(analyses, projections).VisitOperator(*op);
   	op->ResolveOperatorTypes();
   }

   Then GetBinding only needs the scan analysis and scan column index. Keep the existing binding lookup and eligibility checks. The sum(strlen(s)) query in test/optimizer/pushdown/type_pushdown_projection_index.test aborts on the parent ("index 2 within vector of size 1") and returns 100 here.

2. The data.job guard in multi_file_function.hpp can drop the final aggregate row. Vortex emits that row through FinalizeScan, including for empty inputs. With no jobs, neither exhaustion path calls it anymore.

   I think finalization needs its own output metadata. Keep local state for exhausted scans when reader global state exists, and allow finalization without a job.

Hi @lnkuiper. We can reason about (1) later but I don't want to introduce this change here to keep PR small and merge it faster. I've updated (2).

Thanks!