fix(nyc-taxi): make planted freshness defects deterministic

[danielAsaboro]

Summary

The committed nyc_taxi_pipeline.db does not currently contain the two
freshness signals described by its README: the observable stage lag is nine
calendar days rather than three, and no mart row has trip_count = 0.

This change makes the generator deterministic for the sparse 250k-row sample:

• derive the cutoff from the raw pickup business timestamp instead of the
  staging table that is about to be truncated;
• plant the empty load on an existing median mart date instead of assuming
  cutoff - 7 days exists;
• report the actual post-cutoff lag, which can exceed the three-day target when
  sampled dates are sparse;
• regenerate the committed pipeline database and update the README with its
  observed dates; and
• add focused stdlib regression tests for the raw-date cutoff and guaranteed
  empty-load row.

The regenerated committed artifact has:

Signal	Value
Raw max business date	2016-03-10
Staging max business date	2016-03-01
Mart max business date	2016-03-01
Observable stage lag	9 calendar days
Empty-load mart date	2015-01-21
Empty-load trip_count	0

The cutoff target is still raw max minus three calendar days (2016-03-07).
The observable lag is nine days because the committed sample has no rows from
March 2 through March 7. The README now makes that distinction explicit.

The approximately 86.5 MiB generated database is intentionally included in
this patch. This repository already tracks the generated NYC Taxi artifacts,
and its README directs demo users to the committed pipeline database. Updating
only the generator would leave the default download/demo path with the broken
signals reported in #219 and #222. The clean nyc_taxi.db source artifact is
unchanged.

Closes #219.
Closes #222.

Verification

python3 -m unittest -v tests/test_nyc_taxi_create_db.py

Ran 2 tests in 0.010s
OK

cd datasets/nyc-taxi
python3 create_db.py --pipeline-from-existing

Raw data through: 2016-03-10
Staging/mart cutoff target: 2016-03-07 (3 calendar days)
Observed staging max: 2016-03-01 (9 calendar days behind raw)
Empty load: 2015-01-21 shows 0 trips

SELECT MAX(DATE(tpep_pickup_datetime)) FROM raw_trips;
SELECT MAX(trip_date) FROM staging_trips;
SELECT MAX(trip_date) FROM mart_daily_summary;
SELECT trip_date, trip_count
FROM mart_daily_summary
WHERE trip_count = 0;

2016-03-10
2016-03-01
2016-03-01
2015-01-21|0

Scope

This changes only the NYC Taxi dataset generator, its focused regression test,
the generated planted database, and the corresponding README section. It does
not change the clean nyc_taxi.db source artifact or any DataHub service.