feat(drift): add EDDM drift detector

[nikolas-sapa]

Adds a CapyMOA wrapper for MOA's EDDM (Early Drift Detection Method).

EDDM complements DDM. DDM tracks the error rate. EDDM tracks the distance
between consecutive errors. The two often disagree: EDDM is more stable
on noisy streams, DDM reacts faster when the error rate jumps.

In a quick 10-seed check (error rate 1% -> 2% -> 4% -> 8% over 2000
points, changes at 500/1000/1500): EDDM detected all changes with zero
false alarms; DDM also detected all changes but raised 21 false alarms,
including two before the first change. Both are valid choices; they fail
differently.

Notes:

• Thin wrapper over moa.classifiers.core.driftdetection.EDDM.
• EDDM has no tuning parameters in MOA, so the wrapper defines no CLI
  string and get_params() returns {}.
• The docstring example is deterministic: 1000 correct predictions
  followed by 1000 errors. EDDM flags drift at index 1030. Random
  streams produced pre-change false alarms, so the example uses a
  structured stream instead.

Testing:

• uv run invoke fmt
• doctest for the new module passes
• uv run pytest -q tests/test_detector.py (15 passed, 1 skipped)
• uv run pytest -q tests/test_eval_detector.py (11 passed)

Assisted-by: codebuff:deepseek-v4-flash

[tachyonicClock]

That doc test failure appears to be on our end (something probably updated). I will investigate