Checks at once and shards per check default to what the pod's CPUs hold, so a big machine needs no override #2549

Why: #2505 and #2506. The sentinel flags checks.shards=4 and checks.concurrency=6 as away from their defaults (empty = every core, and 3). The owner chose them for the vxrail's 80 cores, so the defaults are what is wrong.

Done looks like:

• [ ] src/core/capacity.ts gains cpuLimit(): the pod's CPU limit from cgroup v2 cpu.max (quota/period), else os.availableParallelism(), in the shape of memoryLimitBytes().
• [ ] The default for checks.shards is min(4, cpus). The default for checks.concurrency is floor(cpus / (shards * 3)), kept between 1 and 16. 80 CPUs give 4 shards and 6 checks; 4 CPUs give 4 and 1. Environment variables and stored values still win.
• [ ] Both help lines say how their default is reached, and the sentinel compares with these defaults.
• [ ] Tests with the CPU count injected cover 4, 16 and 80 CPUs, a stored value overriding the default, and an unreadable cgroup falling back.