fix(train): CpuInfo compared a machine-wide core count against a process-restricted one

[noahgift]

Coverage Nightly is red on main after 7451 passing tests:

thread 'efficiency::device::tests::test_cpu_info_detect' panicked at
  crates/aprender-train/src/efficiency/device/tests.rs:50:5:
assertion failed: cpu.threads >= cpu.cores
test result: FAILED. 7451 passed; 1 failed
make: *** [Makefile:339: coverage] Error 1

threads comes from available_parallelism() — cgroup- and affinity-aware, so it reports what this process may use. cores came from /proc/cpuinfo, which describes the whole machine and honours no restriction. Different denominators; under CPU restriction the machine-wide number is the larger one and the invariant inverts.

Measured on the 24-physical-core runner that failed:

$ awk -F: '/^physical id/{p=$2} /^core id/{print p"-"$2}' /proc/cpuinfo | sort -u | wc -l
24
$ nproc                 # unrestricted
48
$ taskset -c 0,1 nproc  # what a restricted process sees
2

so the assertion became 2 >= 24. Not a flake — with 16 concurrent CI jobs on one box that is the expected reading, and it aborts the run with the rest of the suite unexecuted, so a real regression behind it would be invisible.

usable_cores now clamps the machine-wide physical count to the parallelism actually available, so cores means "physical cores this process can use" — which is what consumers need (estimated_memory_bandwidth_gbps sizes from it).

Mutation check — old expression restored, new tests kept:

detected_physical.unwrap_or_else(|| threads.max(1))  // MUTATION

test usable_cores_never_exceeds_available_parallelism ... FAILED
test usable_cores_upholds_the_cores_le_threads_invariant ... FAILED
assertion `left == right` failed: a process allowed 2 CPUs must not report
  24 usable physical cores

test_cpu_info_detect stayed green under that mutation, because this host is unrestricted — which is exactly why the defect only ever appeared on a loaded CI runner. The new tests drive the reconciliation over its input space instead of over whatever the current machine happens to be.

🤖 Generated with Claude Code

[noahgift]

Superseded by #2449 — this branch is merged verbatim into that batch.

The binding constraint was one ~50-minute workspace-test per PR on one shared box; nine concurrent PRs starved each other into 75-minute step timeouts (every blocked PR this morning classified as CONTENTION, zero merges in 4.5 hours). #2449 lands 24 branches in a single CI run.

Closing rather than leaving open so this PR cannot move #2449's base and force it to re-run. The branch is untouched and this is reopenable if the batch does not land.