Add opt-in interval CSV measurement export

[sankalpsthakur]

Summary

Fixes #467 by writing CSV measurement rows on the same live cadence as API/Prometheus (api_call_interval * measure_power_secs).

• Opt-in csv_run_name (filename, or "auto" / "" → emissions_<run_id>.csv). Default emissions.csv stays final-only unless csv_run_name targets the same file.
• FileOutput.live_out behind enable_live_out — reuses the existing measurement loop instead of ad-hoc flush threads.

Why this matters

Long-running jobs previously had no structured way to inspect intermediate power/emissions without enabling API or Prometheus export. Interval CSV gives operators a file-based audit trail (notebooks, log pipelines, offline analysis) with predictable row cadence and without changing the default single-row emissions.csv contract.

Related (not in scope here): #559 asks whether power columns belong in CSV — this PR follows the existing measurement schema; #448 covers periodic export for other output modes beyond CSV.

Usage

from codecarbon import EmissionsTracker

tracker = EmissionsTracker(
    project_name="Test",
    tracking_mode="machine",
    csv_run_name="emissions_test54.csv",
)
tracker.start()
# ... workload ...
tracker.stop()

Test plan

• tests/output_methods/test_file.py live_out coverage
• tests/test_emissions_tracker.py named / auto / empty / same-as-output_file / default final-only / without CSV method
• Docs: docs/reference/output.md interval CSV note

[sankalpsthakur]

@benoit-cty @mlco2/core-maintainers — implements the #467 approach: opt-in csv_run_name writes measurement rows on the API/Prometheus live cadence. Default emissions.csv stays final-only.

CI is waiting on first-time contributor approval — could a maintainer run the workflows?