[CPBench] Benchmark results Google Sheets writer

[fbonc]

Contributor: Felipe Amaral Bonchristiano (felipea5@illinois.edu)
Contribution Type: New benchmarks/ tool

Description
Adds benchmarks/cpbench/writers, a small CLI for pushing CPBench result JSON files to a Google Sheet so runs accumulate in one place instead of living as scattered local files.

Files to review

• parse_json.py: load_json + flatten, which split a result file into its metadata, aggregated, and per_seed blocks as (header, rows) pairs. Per-seed rows from every seed are concatenated into one block; each row carries its own seed column.
• sheets_writer.py: SheetsWriter.append(worksheet_name, rows, columns). Creates the worksheet if absent, merges new columns into the existing header, and appends.
• cli.py: argparse entrypoint plus the terminal preview rendering.
• __init__.py: package stuff

Usage

# preview: no credentials, no sheet id, nothing written
python -m benchmarks.cpbench.writers.cli --results results/run.json --preview

# append to a sheet
python -m benchmarks.cpbench.writers.cli --results results/run.json \
    --sheet-id 1AbC...xyz --credentials service_account.json

Aggregated and per-seed rows go to separate worksheets, defaulting to "Aggregated Worksheet" and "Per-seed Results" and overridable with --aggregated-worksheet / --per-seed-worksheet. --sheet-id and --credentials are required only when not previewing.

Notes

• Every written row is prefixed with the run's metadata (dataset, task, model, cp_method, timestamp, protocol), so a row identifies itself once many runs share a worksheet.
• Non-scalar metadata is flattened to text; numbers stay numeric.
• Rows are aligned to the sheet's header by column name before appending. New columns are appended, and missing columns are written blank.
• gspread is imported lazily inside _open_worksheet, so it is only needed to actually write — --preview and importing the package work without it.