fix: reject powermetrics on non-Apple-Silicon Macs

[davidberenstein1957]

What changed

ApplePowermetrics._setup_cli() now raises instead of falling through:

• a Mac whose CPU is not Apple Silicon raises SystemError (previously the method returned without assigning self._cli and without raising, so ApplePowermetrics() succeeded on Intel Macs);
• detect_cpu_model() returning None raises the same SystemError rather than an unexpected AttributeError;
• self._cli is assigned on the one path that validated the executable.

_log_values() now uses self._cli instead of a hardcoded "powermetrics", so the validated path is the one actually executed. Two smaller defects in the same argv are fixed alongside it: a stray "" positional argument passed verbatim to powermetrics, and the missing timeout on subprocess.call, which let a hung powermetrics block the measurement thread indefinitely.

Why

Because _setup_cli() never raised on Intel Macs, is_powermetrics_available() returned True whenever a permissive sudo rule existed, and _try_platform_cpu_backend (codecarbon/core/resource_tracker.py:225) routed those machines into _setup_powermetrics, registering an AppleSiliconChip on hardware that has none. With detect_cpu_model() returning None, the resulting AttributeError was swallowed by the broad except Exception in is_powermetrics_available() and reported as powermetrics simply being unavailable, hiding genuine cpuinfo failures.

Note for the changelog: machines that were accidentally using powermetrics on Intel Macs now fall through to Intel Power Gadget or the CPU-load fallback, which is the intended routing but does change their tracking method.

How it was verified

Four new tests in tests/test_powermetrics.py cover the Intel Mac rejection, the None CPU model, the clean argv with a timeout, and the timeout warning path. All four fail on master and pass with this change; the full tests/test_powermetrics.py and tests/test_resource_tracker.py suites pass (51 tests). ruff format --check is clean on both touched files.

Closes#1313

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.45%. Comparing base (eb2cfef) to head (132f230).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@ Coverage Diff @@## master #1333 +/- ##
==========================================
+ Coverage 91.43% 91.45% +0.02% 
==========================================
Files 49 49 Lines 5057 5058 +1 ==========================================
+ Hits 4624 4626 +2 + Misses 433 432 -1 

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[davidberenstein1957]

Closing as folded into #1345. Both are about ApplePowermetrics being dishonest in two ways: it constructs successfully on hardware it cannot measure, and it returns NaN when it measures nothing. Same file, same class, adjacent methods, and they collided textually while both extending tests/test_powermetrics.py, so they are now one 'harden powermetrics' review. The subprocess timeout that was bundled in here has been split out separately as #1397, since it is an unrelated change with its own failure mode.