Implement a log rotation every N benchmarks

We should cut the data files in N elements so that the website do not load a million results and try to render it all at once.

Files could be stored as (i*N)-original-name, so that we can find them easily.

We should also limit the number of elements in the current file to N max. All the excess should go to another file (N*i)-original-name.

The old files will then file up slowly and we will still have enough data in the current files for the graphs.

The issue names no files or tests. Start by locating the benchmark data writer and the website code that reads these files, then trace how the current file is appended and how results are loaded for graphs. Done means files are split at a configurable N-element limit and the website can still find and render the stored benchmark data.

Written by the indexing model from the issue text.