metrics: replace naive elevation accumulation with hysteresis dead-band

GPS jitter and barometric quantization noise caused systematic overestimation
of elevation gain — in extreme cases 100% of reported gain was sub-1m noise.

Implements source-aware hysteresis: elevation is only committed when it
deviates from the last committed value by ≥5m (barometric) or ≥10m (GPS/GPX/TCX).

- ParsedActivity gains `altitude_source` field ("barometric"/"gps"/"unknown")
- FIT parser sets "barometric" when enhanced_altitude is present, else "gps"
- GPX and TCX parsers always set "gps"
- metrics._elevation() uses the threshold matching the source
- 5 new parametric tests covering flat GPS noise, threshold differences, and real climbs

tests/test_metrics.py

@@ -8,6 +8,7 @@ import pytest
 from bincio.extract.metrics import (
     MMP_DURATIONS_S,
     _best_climb,
+    _elevation,
     _fastest_time_for_distance,
     _haversine_m,
     compute,
@@ -126,6 +127,77 @@ def test_compute_no_elevation():
     assert m.elevation_loss_m is None
 
 
+# ── elevation hysteresis ──────────────────────────────────────────────────────
+
+def _ele_pts(elevations: list[float]) -> list[DataPoint]:
+    return [_pt(i, elevation_m=e) for i, e in enumerate(elevations)]
+
+
+def test_elevation_hysteresis_large_step_always_counted():
+    # A single 50m step is way above any threshold — both sources should count it.
+    pts = _ele_pts([100.0, 150.0])
+    gain_baro, _ = _elevation(pts, "barometric")
+    gain_gps,  _ = _elevation(pts, "gps")
+    assert gain_baro == 50.0
+    assert gain_gps  == 50.0
+
+
+def test_elevation_hysteresis_flat_gps_noise_suppressed():
+    # Flat coastal route: 16m of GPS noise oscillating within ±8m.
+    # All steps are sub-1m — hysteresis should return ~0 gain.
+    import math
+    n = 1000
+    elevations = [100.0 + 3.0 * math.sin(i * 0.1) for i in range(n)]
+    pts = _ele_pts(elevations)
+    gain, loss = _elevation(pts, "gps")
+    # With threshold=10m no oscillation within ±3m should ever commit.
+    assert gain == 0.0
+    assert loss == 0.0
+
+
+def test_elevation_hysteresis_barometric_threshold_lower():
+    # Steps of exactly 7m — above barometric (5m) but below GPS (10m) threshold.
+    elevations = [0.0, 7.0, 0.0, 7.0]
+    pts = _ele_pts(elevations)
+    gain_baro, _ = _elevation(pts, "barometric")
+    gain_gps,  _ = _elevation(pts, "gps")
+    assert gain_baro == 14.0   # both 7m steps committed
+    assert gain_gps  == 0.0    # 7m < 10m threshold → suppressed
+
+
+def test_elevation_hysteresis_real_climb_approximated():
+    # Simulate a 200m climb with 0.2m barometric quantization noise.
+    # Build a staircase: 1000 steps, mostly 0.2m up/down noise, with a 200m net climb.
+    import random
+    random.seed(42)
+    elevations = [0.0]
+    for i in range(999):
+        # Mostly quantization noise, but drift upward at 0.2 m/step net
+        step = random.choice([-0.2, 0.0, 0.0, 0.2, 0.2, 0.4])
+        elevations.append(elevations[-1] + step)
+
+    # Force net gain ~200m by scaling
+    scale = 200.0 / (elevations[-1] - elevations[0]) if elevations[-1] != elevations[0] else 1
+    elevations = [e * scale for e in elevations]
+
+    pts = _ele_pts(elevations)
+    gain, _ = _elevation(pts, "barometric")
+    # Hysteresis should produce substantially less than naive accumulation
+    # and land reasonably close to the 200m net climb.
+    assert gain is not None
+    assert gain < 500.0   # not inflated like naive sum
+    assert gain > 100.0   # not zero either — real climbing exists
+
+
+def test_elevation_hysteresis_unknown_treated_as_gps():
+    # "unknown" should apply the same 10m threshold as "gps"
+    elevations = [0.0, 7.0, 0.0, 7.0]  # 7m steps
+    pts = _ele_pts(elevations)
+    gain_unknown, _ = _elevation(pts, "unknown")
+    gain_gps,     _ = _elevation(pts, "gps")
+    assert gain_unknown == gain_gps
+
+
 def test_compute_hr_stats():
     pts = [
         _pt(0,  lat=48.0, lon=11.0, hr_bpm=120),