Enable passing externally-generated witness via command line

[georgwiese]

This PR changes:

• The currently custom-written CSV exporter has been rewritten. It now lives in number::write_polys_csv_file (analogous to number::write_polys_file) and uses the csv crate
  • One feature I removed is that common prefixes are removed. This is for compatibility with the read method; we want to preserve the full column names.
  • I recommend this viewer, which deals nicely with the slightly longer names.
• Added a number::read_polys_csv_file, with a test that the two are compatible.
• Added an --witness-values argument to the pil command that allows us to pass a CSV file with externally-computed witnesses to use during witness generation.

To test:

# Runs the simple sum example on some inputs, stores the trace in a CSV file
cargo run pil test_data/asm/simple_sum.asm --export-csv -i 1,1,1
# Creates a new witness.csv file with only the "main.X_free_value" (see python file below)
python column_subset.py
# Uses witness.csv as externally generated witness. Note that we don't have to provide inputs anymore.
cargo run pil test_data/asm/simple_sum.asm --export-csv --witness-values witness.csv -f

-> column_subset.py

[leonardoalt]

In my new PR I'm changing a lot of these column names as &str to String, so if you want to change more of that stuff feel free. Otherwise I can do it in my PR

[georgwiese]

Ah cool! This would propagate all the way down to witgen, so I'll leave that for another PR. But the write_polys_csv_file() function I introduces already receives a &[(String, Vec<T>)], so you won't have to change that!

[leonardoalt]

yea that's why I thought of that