fix: allow restarting a tracker after stop()

[davidberenstein1957]

What

stop() tore down both schedulers but never cleared _start_time, so a later start() returned early with "Already started tracking". The tracker looked started but was never sampled, and the next stop() wrote a row whose duration covered everything since the original start() — including the idle gap — making that row's duration and emissions_rate wrong.

Changes

• stop() records _paused_at after the final persist, and returns early (with the existing "Tracker already stopped !" warning) if the tracker is already stopped, instead of writing a duplicate row.
• start() rebuilds the schedulers when they were discarded, and on a restart shifts _start_time by the paused interval rather than re-stamping it. The energy accumulators are deliberately not reset on restart, so shifting keeps duration and the accumulators on the same clock: both cover active time only.

Note: a restarted run stays cumulative, and row 2's duration is now the sum of the active phases rather than wall-clock since the first start().

Verification

New test tests/test_emissions_tracker.py::TestCarbonTracker::test_tracker_can_be_restarted_after_stop — start/2s/stop, 2s pause, start/2s/stop — asserts the scheduler exists again after the second start() and that the second row's duration is ~4 s, not ~6 s. It fails on master and passes with this change. Full suite: 627 passed, 21 skipped.

Closes #1328

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.40%. Comparing base (3ec31a0) to head (c8a38b9).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1337      +/-   ##
==========================================
- Coverage   91.43%   91.40%   -0.04%     
==========================================
  Files          49       49              
  Lines        5057     5060       +3     
==========================================
+ Hits         4624     4625       +1     
- Misses        433      435       +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.

[davidberenstein1957]

Closing as folded into #1336. Both PRs were about start() and stop() disagreeing on whether the tracker has run, and they had already converged: #1336 introduces _stopped_at as the canonical lifecycle flag, while this PR was inferring the same state from _scheduler is None. Two flags for one fact. The refusal now reads 'if self._stopped_at is not None' on #1336, so it is one flag and three guards in a single review. The honest-error change and its test came across unchanged.