I think that it is good enough for now to merge this

All tests are adjusted accordingly and swc_read_cells() is now removed.

Notable difference: parse_record() now throws only an swc_parse_error.
In case of an invalid_argument raised by the cell_record constructor,
this is caught and rethrown as a swc_parse_error with line information
attached.

Begin iterator resets the input stream.

Adde `-g' flag in compilation to allow debugging.

Use of cell record ranges for hiding the internals of file reading. This is
a proof-of-concept for a first peer review and discussion. Check usage
paradigm in corresponding test. Now we could easily support stateful parsing
reporting context of problematic lines.

Cell record cleaning + minor other improvements

Moved SWC parser's and cell record's implementation details to a
separate file. Also added braces around single-line control/loop blocks.

Read in cell records and do a basic cleaning:
  1. Remove duplicates.
     Only the first occurance is considered.
  2. Sort the records if not sorted.
  3. Ignore subsequent trees if more than one are specified.
  4. Renumber records if numbering is not contiguous.

This exception is thrown when a parse error occurs, instead of
std::runtime_error and std::logical_error that used to be thrown before.

The will be useful for sorting cell records. Equality and ordering is
defined solely on records IDs.

Refactor the project to have source and tests in different paths and use cmake.

Swc io

The line buffer is no more a pure character array but a standard string.
For reading lines from the input stream we use the higher level
std::getline() function which takes care of overflow issues.

Side improvements: String comment prefixes are now supported.

Other changes: added `-pedantic' to the compilation flags, to force
strict C++11 compliance.