fix: handle zero-duration clusters in diff_sessions (closes #79)

[conorluddy]

Closes #79.

Problem

diff_sessions() had a divide-by-zero guard that did a bare continue:

if ca.max_duration_ms == 0:
    continue

A zero-duration cluster in session A simply vanished from the diff. Worse, a real regression — a previously-silent cluster (0 ms) that started hanging (e.g. 800 ms in B) — was dropped entirely instead of being flagged as drift.

Fix

Replace with explicit 4-way branching:

Case	Result
A == 0, B == 0	stable++
A == 0, B > 0	drift, delta_pct = float("inf") (worsened, "new signal")
A > 0, B == 0	drift, delta_pct = -100.0 (fully improved)
A > 0, B > 0	existing percentage logic

The inf delta is special-cased in format_diff to render as new rather than +inf%.

Tests

5 new in tests/test_diff.py:

• test_diff_zero_to_zero_counts_as_stable
• test_diff_zero_to_nonzero_is_drift_with_inf_delta
• test_diff_nonzero_to_zero_is_drift_with_minus_one_hundred
• test_diff_nonzero_unchanged_still_uses_threshold (regression guard)
• test_format_diff_renders_inf_delta_as_new

A small helper _zero_dur_summary() builds SessionSummary instances with controlled max_duration_ms per cluster — keeps the test isolated from the event-aggregation pipeline.

Verification

ruff check: All checks passed
black --check: 42 files would be left unchanged
python -m pytest tests/: 93 passed (was 88)
pytest tests/: 93 passed

[conorluddy]

Superseded by #90 — same commits combined into one PR.