fix(projection): align projection with ADR 0002

Use lifetime written bytes from the current segment's newest published
sample in the headline formula, select the Percentage-Used-implied
baseline after two increments in the segment (otherwise show the
too-coarse fact), and judge horizon agreement across existing horizons.
Make the vacuous projection tests assert unconditionally and add
coverage for implied-baseline gating and single-horizon agreement.

tests/test_acceptance_sweep.py

@@ -98,8 +98,8 @@ def _insert_day(conn, day, bw=1024*1024*100, coverage=0.95, samples=24):
 def _insert_sample(conn, ts, pu=5):
     conn.execute(
         "INSERT INTO samples (ts, device, data_units_written, data_units_read, "
-        "percentage_used, bytes_written, bytes_read, power_on_hours) "
-        "VALUES (?, ?, ?, ?, ?, ?, ?, ?)",
+        "percentage_used, bytes_written, bytes_read, power_on_hours, segment_id) "
+        "VALUES (?, ?, ?, ?, ?, ?, ?, ?, 1)",
         (ts, "/dev/nvme0n1", 1000000, 500000, pu, 512000000000, 256000000000, 8765),
     )
     conn.commit()