PhParser: allow for pattern initialization

[bastonero]

The ph.x should be parallelized by setting in the input parameters start_irr and last_irr to 0. This allows the program to exit smoothly and it further avoids to wait for the rewriting of the wavefunctions, which can be a rather long and intensive IO operation, not really suited for initialization runs.

The parser is then adjusted to account for this option, as for some reason the line having JOB DONE is not printed in such cases. A simple specialized parser is also added to store the number of q-points and their values, which can be later on used to parallelize over q-points by specifying last_q and start_q.

PS: this also avoids the creation of the aiida.EXIT file, which triggers the program to stop.

Note: this is also the recommended way from the official webpage. To report the statements in case the link will be broken in the future:

NB: The program ph.x writes on the tmp_dir/_ph0/{prefix}.phsave directory
a file for each representation of each q point. This file is called
dynmat.#iq.#irr.xml where #iq is the number of the q point and #irr
is the number of the representation. These files contain the
contribution to the dynamical matrix of the irr representation for the
iq point.

If [recover](https://www.quantum-espresso.org/Doc/INPUT_PH.html#recover)=.true. ph.x does not recalculate the
representations already saved in the tmp_dir/_ph0/{prefix}.phsave
directory.  Moreover ph.x writes on the files patterns.#iq.xml in the
tmp_dir/_ph0/{prefix}.phsave directory the displacement patterns that it
is using. If [recover](https://www.quantum-espresso.org/Doc/INPUT_PH.html#recover)=.true. ph.x does not recalculate the
displacement patterns found in the tmp_dir/_ph0/{prefix}.phsave directory.

This mechanism allows:

  1) To recover part of the ph.x calculation even if the recover file
     or files are corrupted. You just remove the _ph0/{prefix}.recover
     files from the tmp_dir directory. You can also remove all the _ph0
     files and keep only the _ph0/{prefix}.phsave directory.

  2) To split a phonon calculation into several jobs for different
     machines (or set of nodes). Each machine calculates a subset of
     the representations and saves its dynmat.#iq.#irr.xml files on
     its tmp_dir/_ph0/{prefix}.phsave directory. Then you collect all the
     dynmat.#iq.#irr.xml files in one directory and run ph.x to
     collect all the dynamical matrices and diagonalize them.

NB: To split the q points in different machines, use the input
variables start_q and last_q. To split the irreducible
representations, use the input variables [start_irr](https://www.quantum-espresso.org/Doc/INPUT_PH.html#start_irr), [last_irr](https://www.quantum-espresso.org/Doc/INPUT_PH.html#last_irr). Please
note that different machines will use, in general, different
displacement patterns and it is not possible to recollect partial
dynamical matrices generated with different displacement patterns.  A
calculation split into different machines will run as follows: A
preparatory run of ph.x with [start_irr](https://www.quantum-espresso.org/Doc/INPUT_PH.html#start_irr)=0, [last_irr](https://www.quantum-espresso.org/Doc/INPUT_PH.html#last_irr)=0 produces the sets
of displacement patterns and save them on the patterns.#iq.xml files.
These files are copied in all the tmp_dir/_ph0/{prefix}.phsave directories
of the machines where you plan to run ph.x. ph.x is run in different
machines with complementary sets of start_q, last_q, [start_irr](https://www.quantum-espresso.org/Doc/INPUT_PH.html#start_irr) and
[last_irr](https://www.quantum-espresso.org/Doc/INPUT_PH.html#last_irr) variables.  All the files dynmat.#iq.#irr.xml are
collected on a single tmp_dir/_ph0/{prefix}.phsave directory (remember to
collect also dynmat.#iq.0.xml).  A final run of ph.x in this
machine collects all the data contained in the files and diagonalizes
the dynamical matrices.  This is done requesting a complete dispersion
calculation without using start_q, last_q, [start_irr](https://www.quantum-espresso.org/Doc/INPUT_PH.html#start_irr), or [last_irr](https://www.quantum-espresso.org/Doc/INPUT_PH.html#last_irr).
See an example in examples/GRID_example.

On parallel machines the q point and the irreps calculations can be split
automatically using the -nimage flag. See the phonon user guide for further
information.

[sphuber]

Thanks @bastonero . Just the one test is failing. If you fix that, I will merge

[bastonero]

@sphuber i just added the small .gitkeep file, and finally the PR is not anymore blocked by some older PR re docs or similar.

[sphuber]

Thanks a lot @bastonero