Benchmark files: read the encoding settings from the file itself before labelling results 'default config'

Situation: a large test file in a compressed, block-based format had been used for weeks of read and memory benchmarks, all quoted as if it used the format's default settings. When someone finally read its header and per-block layout with the format's own dump tool, it turned out to use a different compression level, a smaller block size and no precomputed summaries. The script that built it had simply left out the level flag, so the packer's default at that commit applied, and that default differed from the documented one.

Practice: - For every benchmark input, record the settings read from the file (block size, compression level if stored, optional sections present) plus the file's sha256, and put them in every result row. - If a setting such as the compression level isn't stored in the file, recover it from the exact build command and the packer's default at that commit, and say that's how you got it. - Count rows from tool output carefully. A header line counted as data gave an off-by-one block count. - When the file turns out to be non-default, relabel every result derived from it, and plan a default-config encode before claiming default-config numbers.