[tanmay-9]

Since #305, the QLever binaries sample their own RSS memory and CPU usage and write the resource-usage log themselves, so the Python-side sampler was removed. The other engines of qeval do not write such a log, so they need a sampler on the Python side. The plot code so far also only described a QLever index build, since the shaded phase bands and the subtitle come from the QLever index log.

With this change, qeval/resource_monitor.py provides that sampler. It wraps the index command of an engine and writes a log with the columns that the plot needs (elapsed_s, rss, cpu_percent), by following the index process natively or with docker stats or podman stats in a container. The plot code is shared by all engines. Only the shaded bands and the subtitle differ per engine, and an engine passes its own engine_overlay and engine_subtitle functions to render_usage_plot_if_installed in the new qlever/resource_usage/plot_util.py. Without them, the plot describes a QLever index build, so qlever index works as before. plot_util.py can be imported without matplotlib and numpy, so the plotting libraries stay optional for every engine.

The log and the plot files of a non-QLever engine are prefixed <dataset>.<engine>, so that it is clear which engine a file belongs to. QLever keeps the plain <dataset> prefix, since its own binaries write that log.

On the side, the two helpers in qlever/util.py that only the sampler uses moved along with it to qeval, and parse_git_hash moved next to its only caller in usage_plot.py. The shared test fixtures, including the guard against signals to unrelated processes, moved to test/conftest.py, so that they also apply to the tests of qeval.

NOTE: No engine uses the sampler yet. Oxigraph will be the first one, in #328.