save_to_* DeprecationWarning has the wrong stacklevel and never reaches users #1323

The deprecation warning for the save_to_* constructor parameters is emitted with a stacklevel that points inside CodeCarbon itself, so Python's default ignore::DeprecationWarning filter drops it for every caller that is not a top-level __main__ script. Users who pass save_to_file=True from a library module are never told to migrate to output_methods=.

Reproduction

# mylib.py
from codecarbon import EmissionsTracker
tracker = EmissionsTracker(save_to_file=True)   # deprecated

python -c "import mylib" prints nothing. Under warnings.simplefilter("always") the warning is recorded, but its filename is codecarbon/emissions_tracker.py, not the calling module.

Root cause

codecarbon/emissions_tracker.py:227 warns with stacklevel=2 from _resolve_output_methods, which is called from BaseEmissionsTracker.__init__ (emissions_tracker.py:569). Frame 2 is therefore __init__ itself. A fixed stacklevel=3 is not sufficient either: OfflineEmissionsTracker.__init__ (emissions_tracker.py:1313) adds another frame, so the correct level depends on the entry point.

Secondary: CodeCarbon's own CLI passes deprecated flags, so once the attribution is corrected the CLI warns users about a choice the CLI made — codecarbon/cli/monitor.py:71 (save_to_logger=False, which is also the default and therefore purely redundant), codecarbon/cli/main.py:425 (save_to_api=api) and codecarbon/cli/main.py:476 (save_to_file=False). Today these are hidden by the same default filter, but they hard-fail under -W error::DeprecationWarning or a suite with filterwarnings = error.

Expected vs actual

Expected: the warning is attributed to the user code that constructed the tracker, so it surfaces under the default filter, and CodeCarbon's own CLI does not trigger it.

Actual: the warning is attributed to CodeCarbon internals, is silently dropped for most users, and the CLI trips its own deprecation check.