Add `table-to-csv-file` to allow writing tables to CSVs

[dbp]

An assignment workflow I'd like to be able to support:

• In preprocess.arr:

  include csv
  raw-table = load-table: ...
    source: csv-table-url("...")
  end
  
  processed-table = ...
   
  table-to-csv-file(processed-table, "processed.csv")

• In analyze.arr:

  include csv
  processed-table = load-table: ...
    source: csv-table-file("processed.csv")
  end
  ...

The motivation for this is to allow students to start from actual data (e.g., I have a 76MB property assessment database for Boston that I'd like to use), which works, but is slow (in our tests, took about 20-30 seconds to load on a pretty new laptop), too slow to be loaded and filtered on every run.

Alternately, rather than processed.csv, it could be test-data.csv, with the idea that they first identify a subset to use for testing, and use that for their development of analyze.arr, and then at the end, once it works, replace test-data.csv with the raw url.

Obviously, we could do the preprocess.arr steps offline and just present them with processed.csv but, given they know all the skills to implement it, and in the VSCode/Github setting, having multiple files is not a challenge, this minor change would allow it to be much more realistic (and easier for them to, e.g., adapt to other scenarios).

(I know this workflow could probably be even better supported by the jupiter-esque UI you have in development, but I think that's too much of a change for me, right now).

[jpolitz]

This is a great idea. I think the principled approach here is serializers that are the complement of sanitizers, since arbitrary values can be in table cells.

I think the MVP approach is to have default serializers for num, bool, string that are guaranteed to round-trip through csv-table-url as much as is feasible, and error on other table values.