Fix pre-existing test failures in test_writer and test_metrics

test_writer: _dummy_metrics() and test_build_summary_required_fields were
missing np_power_w=None after the field was added to ComputedMetrics.

test_metrics: the leading-zero elevation heuristic fired on a single 0.0
start value, incorrectly skipping the first legitimate elevation step.
Guard now requires at least 2 consecutive near-zero leading values before
activating the Apple Watch lock-acquisition workaround.

bincio/extract/metrics.py

@@ -377,16 +377,21 @@ def _elevation(
 
     # Some devices (e.g. Apple Watch) record exactly 0.0 for the initial samples
     # while waiting for barometric/GPS lock, then jump to the real altitude.
-    # Detect this by checking for a leading near-zero run followed by a large
-    # jump: skip those zeros and seed the accumulator from the first real value.
-    # Safety: if the activity genuinely stays near sea level, no value in the
-    # series will exceed `threshold`, so `start` stays 0 — unchanged behaviour.
+    # Only activate when there are at least 2 consecutive near-zero leading
+    # values — a single 0.0 is a legitimate sea-level starting point.
     start = 0
     if abs(elevations[0]) < 0.5:
-        for i, e in enumerate(elevations):
-            if abs(e) > threshold:
-                start = i
+        n_leading = 0
+        for e in elevations:
+            if abs(e) < 0.5:
+                n_leading += 1
+            else:
                 break
+        if n_leading > 1:
+            for i, e in enumerate(elevations):
+                if abs(e) > threshold:
+                    start = i
+                    break
 
     gain = loss = 0.0
     committed = elevations[start]