fix(cli): detect macOS logical CPU count via sysctl, not just nproc

[roshangupta00750]

Problem

ods-cli's _get_host_logical_cpus() reports 1 logical CPU on macOS:

cores=$(nproc 2>/dev/null || grep -c '^processor' /proc/cpuinfo 2>/dev/null || echo "1")

nproc and /proc/cpuinfo are Linux-only; on macOS both fail and the chain falls through to echo "1". That value backs _get_docker_available_cpus()'s fallback (used when docker info --format {{.NCPU}} is unavailable), and the result caps the per-service CPU budget — so an Apple-silicon host provisions llama-server and the bundled services for a single core.

Reproduced on macOS:

( set -- version; source ./ods-cli >/dev/null 2>&1; _get_host_logical_cpus )  # 1
sysctl -n hw.logicalcpu                                                       # 12

Fix

Probe sysctl -n hw.logicalcpu (macOS/BSD) between the two Linux sources. Order is preserved on Linux — nproc still wins first — so only the previously-broken macOS path changes.

Sibling of the same nproc-only gap being fixed for installers/lib/detection.sh in #5564 / #2827; those PRs do not touch ods-cli.

Validation

Adds tests/test-ods-cli-host-cpu-fallback.sh (wired into make test).

• Before: on macOS _get_host_logical_cpus → 1; the new test's stubbed case (nproc unavailable, sysctl→7) returns the Linux /proc/cpuinfo count instead of 7 → FAIL.
• After: macOS returns its true count (12 on the test host); the stubbed sysctl case returns 7 on both macOS and the Linux VM.
• No Linux change: with nproc present the result is unchanged (VM: 4 = nproc).
• Teeth: reverting the ods-cli change makes the test fail again on both platforms.
• bash -n and shellcheck -S error ods-cli clean.

Closes 5641