Closes #4.

*Example*

>> a = dseries(randn(4,5),'2000Q1')

a is a dseries object:

       | Variable_1 | Variable_2 | Variable_3 | Variable_4 | Variable_5
2000Q1 | 2.4245     | -1.036     | -0.87587   | -0.57001   | -1.6989
2000Q2 | 0.9594     | 1.8779     | 0.31995    | -1.0257    | 0.6076
2000Q3 | -0.31577   | 0.9407     | -0.55829   | -0.90875   | -0.1178
2000Q4 | 0.42862    | 0.78735    | -0.31143   | -0.2099    | 0.69916

>> a.Variable_3>0

ans =

  4x1 logical array

   0
   1
   0
   0

>> a(a.Variable_3>0)

ans is a dseries object:

       | Variable_1 | Variable_2 | Variable_3 | Variable_4 | Variable_5
2000Q2 | 0.9594     | 1.8779     | 0.31995    | -1.0257    | 0.6076

Ref #4.

If ts is a dseries object with T periods and N variables, then we can
instantiate a new dseries object from ts by extracting periods or
variables. Periods are selected by passing a vector of logicals with T
elements. Variables are selected by passing a vector of logicals with
N elements. If T==N, the convention is to pass a row vector to select
variables and a colum vector to select periods.

*Example*

>> ts = dseries(randn(10,5), '2000Q1')

ts is a dseries object:

       | Variable_1 | Variable_2 | Variable_3 | Variable_4 | Variable_5
2000Q1 | -0.53201   | -1.0642    | 0.39189    | -1.0667    | -0.73417
2000Q2 | 1.6821     | 1.6035     | -1.2507    | 0.93373    | -0.030814
2000Q3 | -0.87573   | 1.2347     | -0.94796   | 0.35032    | 0.23235
2000Q4 | -0.48382   | -0.22963   | -0.74111   | -0.029006  | 0.42639
2001Q1 | -0.712     | -1.5062    | -0.50782   | 0.18245    | -0.37281
2001Q2 | -1.1742    | -0.44463   | -0.32058   | -1.5651    | -0.23645
2001Q3 | -0.19224   | -0.15594   | 0.012469   | -0.084539  | 2.0237
2001Q4 | -0.27407   | 0.27607    | -3.0292    | 1.6039     | -2.2584
2002Q1 | 1.5301     | -0.26116   | -0.45701   | 0.098348   | 2.2294
2002Q2 | -0.24902   | 0.44342    | 1.2424     | 0.041374   | 0.33756

>> ts(mean(ts)>0)

ans is a dseries object:

       | Variable_4 | Variable_5
2000Q1 | -1.0667    | -0.73417
2000Q2 | 0.93373    | -0.030814
2000Q3 | 0.35032    | 0.23235
2000Q4 | -0.029006  | 0.42639
2001Q1 | 0.18245    | -0.37281
2001Q2 | -1.5651    | -0.23645
2001Q3 | -0.084539  | 2.0237
2001Q4 | 1.6039     | -2.2584
2002Q1 | 0.098348   | 2.2294
2002Q2 | 0.041374   | 0.33756

>> ts(ts.dates>=dates('2001Q3'))

ans is a dseries object:

       | Variable_1 | Variable_2 | Variable_3 | Variable_4 | Variable_5
2001Q3 | -0.19224   | -0.15594   | 0.012469   | -0.084539  | 2.0237
2001Q4 | -0.27407   | 0.27607    | -3.0292    | 1.6039     | -2.2584
2002Q1 | 1.5301     | -0.26116   | -0.45701   | 0.098348   | 2.2294
2002Q2 | -0.24902   | 0.44342    | 1.2424     | 0.041374   | 0.33756

Concatenation was failing if the empty object had no frequency set.

Closes #46

Closes: #47

The returned argument is the first input argument.

Conversion to yearly and bi-annual frequencies.

Returned year was wrong for yearly frequency.

The dates2S routine was plain wrong.

H (half year) is still allowed.

Bug introduced in 7bb6a68f when changing the implementation of the
@dates class.

Also:

 - added unit tests,
 - added conversion to bi annual dates.

For all frequencies, counts the number of subperiods since year 0.

First argument, specifying the frequency, can be an integer *or* a character.

This is not necessary since ordering in INFO should not matter, also
this option is not available in Octave 5.x.

The round function in octave does not admit two arguments (the second
being the rounding precision) as in recent versions of Matlab. The same
trick should apply to old Matlab versions, but I do not know when
Mathworks introduced the second argument… A simple solution could be
to always use this trick, and never use the second argument.

This reverts commit 933c51d5.