feat(cli): live local dashboard for monitor

[davidberenstein1957]

Closes #1348

What it adds

codecarbon monitor --ui serves a live view of the current run on http://127.0.0.1:8050: current power, cumulative gCO2eq, energy, elapsed time, a CPU/GPU/RAM power chart over the last samples, the detected hardware, and per-task rows when tasks are used.

It is a stdlib ThreadingHTTPServer on a daemon thread plus one self-contained HTML file. No new dependency, no build step, no database, no CDN, no network access — it works air-gapped and does not need the carbonboard extra or carbonserver.

User-facing surface

• codecarbon monitor --ui [--ui-port 8050] [--ui-host 127.0.0.1], also on the wrapped form (codecarbon monitor --ui -- python train.py).
• codecarbon.viz.live.LiveDashboardOutput, usable like any other output handler: EmissionsTracker(output_handlers=[LiveDashboardOutput()]).
• Routes: / (page), /data (JSON), /health.
• Loopback by default; a non-loopback bind logs a warning, since the server is unauthenticated by design. A busy port logs an error and monitoring continues without the UI.
• Docs: new section in docs/how-to/visualize.md, options in docs/reference/cli.md.

One change outside the new files

live_out was only called every api_call_interval measures (5 minutes with the defaults), which is far too coarse for a live view. Handlers can now set live_out_every_measure = True to be called on every measurement instead, receiving the total with delta=None — the delta is deliberately not computed there, since _compute_emissions_delta consumes state used by the periodic call. Existing handlers are unaffected: the attribute defaults to False and is read with getattr, so duck-typed custom handlers keep working.

Verified

• uv run pytest tests/test_live_dashboard.py -q — 8 tests: bounded history, /data payload, / and 404, /health, concurrent read/write returning valid JSON, port released on exit(), busy port stays inert, and the tracker feeding a sample per measure.
• No regression in tests/test_custom_handler.py, tests/test_emissions_tracker.py, tests/cli, tests/test_docs_examples.py, tests/test_package_integrity.py.
• Manual run: codecarbon monitor --ui --offline --country-iso-code FRA --measure-power-secs 3, samples appearing every 3 s and the page rendering.
• black --check and ruff check scoped to the touched files (remaining ruff hits in cli/main.py are pre-existing).

Deliberately left out

• Reading existing emissions.csv history — that is carbonboard's job; this only shows the current run.
• Auth, TLS, persistence, websockets, config-file keys (ui, ui_port, ui_host). Flags seem enough for an interactive foreground command; happy to add the config keys if wanted.
• A docs screenshot, which I cannot produce here — worth adding before merge.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.58%. Comparing base (065d0e6) to head (61a0472).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1365      +/-   ##
==========================================
+ Coverage   91.39%   91.58%   +0.18%     
==========================================
  Files          49       50       +1     
  Lines        5056     5145      +89     
==========================================
+ Hits         4621     4712      +91     
+ Misses        435      433       -2     

☔ 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.