metadata: add cpu_quota (Linux cgroup CPU bandwidth limit)

[manunicholasjacob]

A benchmark run inside a container or a systemd unit with a CPU limit (docker run --cpus=2, a Kubernetes CPU limit, systemd CPUQuota=) gets throttled by the kernel once it uses its quota in a period, but nothing in the metadata shows it: cpu_count is still the host count and cpu_affinity is not set, because a bandwidth limit does not change the affinity mask. Two results from the same machine can then differ for a reason compare_to cannot show.

This PR adds a cpu_quota metadata entry on Linux, read from the process's own cgroup:

• cgroup v2: cpu.max of the process cgroup (from /proc/self/cgroup) and of each parent, keeping the smallest limit, since a parent limit also applies.
• cgroup v1: cpu.cfs_quota_us / cpu.cfs_period_us, falling back to the mount root because the cgroup path is not visible inside a container.

The value is formatted like the other CPU entries, e.g. 2 CPUs or 1.5 CPUs, and the key is only set when a limit is set, like cpu_affinity. The new entry is documented in doc/api.rst.

Tests: three unit tests with mocked /proc and /sys files (v2 leaf, tighter parent and container root, v1 in a container and -1 for no limit, no limit at all). ruff check (0.15.22 as pinned in pre-commit) passes. The rest of the suite gives the same result with and without this change on my machine.

Checked on Linux (WSL2, 20 logical CPUs, cgroup v2), Python 3.14 and 3.11:

run	cpu_count	cpu_affinity	cpu_quota
no limit	20	(not set)	(not set)
systemd-run --user --scope -p CPUQuota=150%	20	(not set)	1.5 CPUs
docker run --cpus=2	20	(not set)	2 CPUs

I used an AI assistant (Claude) to help write this change. I ran the checks above myself.