The three-series case (milestone 2, the option-(b) measurement).

Data: Lutkepohl (2005), data set E1 -- West German fixed investment, disposable
income and consumption expenditures, billions of DM, quarterly, seasonally
adjusted, 1960Q1-1982Q4 (92 observations); source Deutsche Bundesbank. Taken
byte for byte from atsw-gui engines/drvec/datasets/lutkepohl/e1.dat (SHA-256
c4fdef149249cea596657c42713ce2f41df160d23c29ed1b9b59e2ff10fc0d97, listed in
engines/drvec/datasets/MANIFEST.md).

e1invest.inp, e1income.inp, e1cons.inp: one fue specification per column,
written by a script: frequency 4, start 1960 Q1, no deterministic variables,
no ARMA, the mean free (0.0 1), Box-Cox lambda 0 (logs), d = 1, D = 0,
rescaling factor 1. Each was estimated with fue 0.1.17 (`python -m fue
e1<name> eml`), which wrote the ../e1<name>.pre files: the .pre is written by
the program that estimated it, which is the format's rule.

What drvec reads from them is the series, the calendar, lambda and the
rescaling factor (docs/PORT.md §7); the fue model (a random walk with drift in
logs) is not read on the .pre route.
