Fixed the array representation. Before this fix, the arrays of values were
represented as "[a, b, c]", which could lead to tremendous bugs due to
ambiguity (the example array could have three values, "a", "b" and "c", or only
one, "a, b, c", or other combinations). The fixed notation solves that
ambiguity, because now the arrays are represented in the form: ["a", "b", "c"]

Now the user cannot only specify if he wants to represent the arrays flattened
or compressed in the CSV file, but also which depth of compression wih an
integer. A level of 0 would equal no compression, a level 1 equals the old
compression mode (and the current UPV-compatible mode too), that is, to just
unnest 1 level of arrays, for example:

house[0].dweller[0]: "Ana"
house[0].dweller[1]: "Roberto"
house[1].dweller[0]: "Marina"

Would become:
house[0].dweller: "[Ana, Roberto]"
house[1].dweller: "[Marina]"

Now, it is also possible to select a compression of depth 2 or more, which
would provide the followint output:

house.dweller: "[[Ana, Roberto], [Marina]]"

Some minor bugs fixed and tests adapted to the new feature.
This is the state of the program desired in the beginning of the development,
and this version (with possible minor fixes) is the one expected for the 1.0.0
release.

Since the program has reached a pretty stable stage (to the level provided by
automatic and manual testing done by the developer) and with a
reasonably good functionality in comparison to its initial goals, the
changes are dropped to master.
New release version boys!!

Fixed some bugfixes related to the changes introduced at 0.8.0. E.g., in 0.8.0
one could not flatten the split directory of the same collection, because the
collection was removed before (the overwriting was performed over the top
directory of the collection, and thus the input for the operation was removed
before beginning to flatten). Tests have been adapted too

This patch includes a big reestructuration of the directory hierarchy produced
by the program. The three standard folders for output (split\_tweets,
flattened\_tweets and csv\_converted\_tweets) cease to exist. Now the only
preexistent folder for output is "output" itself. Everytime a collection is
split, flattened or converted, a new output folder with the name of that
collection is created inside "output". These collections can have up to three
subdirectories called "Split", "Flat" and "CSV", every one of which contains
the type of files indicated by its name.

The possibility of discarding or keeping the intermediate files (split and/or
flattened JSONs) has been added as an invocation argument (-k / --keep-files).
Before this version these files were always kept. Now they are discarded by
default and the -k options permits to keep them.
Before this version, the array compression and remove dollar modes didn't apply
when there wasn't an include filter (that is, either with exclude filter or
without any filters at all). This has been fixed and now any combination of
filters and modes should work correctly.

Before this version, the array compression and remove dollar modes didn't apply
when there wasn't an include filter (that is, either with exclude filter or
without any filters at all). This has been fixed and now any combination of
filters and modes should work correctly.

Now the user will be prompted whenever a collection with the same path
than an already existing one is going to be generated. The user will be
able to choose between overwriting or aborting the execution.

Filters are no longer hardcoded. Now, if user doesn't specify anything
about filters, none will be applied. The user can specify the path to an
exclude or an include filter using the -e/--exclude and the -i/--include
command-line options. Furthermore, the user can specify
-f/--default-filters to use the default filters included in data/filters