1. Calculate columns using expressions that reference other columns
2. If/then logic to build rules around calculations
3. Aggregates
4. Lookup values from other columns or columns from other CSV files that have been loaded
5. Joins
6. More string manipulation functions (e.g. extract substring, trim left, trim right, regex operations)
Make full use of whatever awesome multifile read/edit the software presumably has already and then maybe go a little beyond. The "configuration" csv might have some hashbang equivalent defining a line offset for the header that isn't mapped into input/output files (perhaps a concept of "negative line numbers" for project properties?), contains references to those peer files and maybe activates project options like "display peer file contents in cells that are otherwise empty" if you want your working surface to remain "2d".
The argument to have this add-on (I'm not sure that it should be part of the product or a separate extension) would be that it would be a shame to learn all the UI ergonomics of the plain reader/editor and then not be able to leverage them for some calculations as well.
I would actually be wary of trying to make this too much like a spreadsheet, even in appearance. You'll be pulled into that strange attractor, but you can't hope to compete in that space. Whatever oxygen isn't consumed by Excel itself is long gone between LibreOffice and online spreadsheets.
Another possibility would be integrating with SQLite via CSV export and giving people an SQLite console: https://www.sqlitetutorial.net/sqlite-import-csv/
I recently was given a pile of hand-edited CSV files (a one off data transfer between two systems). The original export was missing one column, so being able to merge that in would have been very useful.
I'm sure with Pandas if you know the magic incantation it's easy to do (join two files on the ID column, and add column X from file 2 to file 1) but I wrote my own script for these files.
In others I've resorted to opening the CSV in Excel, doing an XLOOKUP and then fixing up the mess Excel created in Modern CSV afterwards.
(To the sibling commenter saying you can do it with SQL: yes and I'd love to. But how do you pipeline that with multiple CSVs that may have slightly different column names? If done manually it seems quicker to do it my way)