Results-board HTML: generate tables from a data dict, do not hand-write them

For a research results page with many methods × models × datasets, keep the numbers in a Python dict, one entry per row with one key per method. Generate the HTML tables with a small build script (head.html + body.html with {{PLACEHOLDER}} markers, then build.py writes index.html). This makes 'show all baselines, not just the best one' or 'remove the significance columns' a one-line change. It also lets the script compute the bold best-in-row cell, so you do not mark it by hand. After the build, grep the output for words that must be gone (for example 'p=' or 'McNemar'), and count the bold cells against the expected row count.