Preprocessor: no longer use boost::graph::GraphvizDigraph class, removed in Boost 1.44 (closes #155)

- Bug correction during the block decomposition: the models' graph was ill oriented before the search of strong components

added choice of parameter set to shock_decomposition. This corrects a bug when shock_decomposition is requested after Metropolis iterations.

fixed bug that would force node sharing when two or more expectation operators had the same expression but different information sets

Bugfix at order 3 (hopefully was not affecting final results, thanks to Pablo Winant for spotting this):
- in the 3rd derivatives matrix, among symmetric elements, the first one had
the right values but the following ones were set to zero
- moreover, the k-order DLL was trying to add all the symmetric elements in the
folded tensor, instead of only keeping one value among all the symmetric ones
- hopefully, Ondra's tensor library was (silently) refusing to add symmetric
elements after the first (and right) value had been added
- so the final result was correct

- Correction of a bug in the Jacobian Matrix computation for a non block-decomposed model using bytecode

and corrects the computation of the number of column in the jacobian matrices in a block decomposed model without bytecode
- Identical corrections for models using bytecode without block decomposition

Add the missing dr_block m-file

Arbirtrary expressions after the "values" keywords must now be enclosed
within parentheses; consider the following example:
<code>
periods 1:2;
values -1 -2;
</code>

In the previous syntax, this was interpreted by the preprocessor as a shock of
value -1-2 = -3 for periods 1 and 2, which is clearly not the intent of the
user; with the new syntax, this will be rejected (too many values compared to
the number of ranges).

Also note that now commas are no longer required between arbitrary expressions,
since the parentheses are sufficient for separating them.

Preprocessor: max lead/lags in the model are no longer affected by lead/lags on parameters (closes #153)

- Add wrapper needed to compute the steady-state using bytecode and solve_algo = 0, ..., 4

- fail when user tries to initialize a variable at a period which is not in the model, and make the error message explicit
- correctly handle exogenous variables in stochastic mode, which are also substituted with an aux var
- added test files

Preprocessor: fix possible bug in histval (need to limit search of aux vars to aux vars for lagged vars, otherwise we could have a wrong match with some other kind)

- Extends the algorithms compatible with bytecode to compute the  steady-state (ticket #11). The new values for solve_algo are:
   * 0: fsolve
   * 1: solve1
   * 2, 4: solve1 + block decomposition
   * 3: csolve
   * 5: LU decomposition with UMFPack (method handling sparse matrix in Matlab)
   * 6: GMRES
   * 7: BiCGStab
   * 8: bytecode own solver (use Gaussian elimination + sparse matrix)

- Bytecode can now evaluate a specific block instead of the overall blocks (new bytecode's option 'Block = block_number')

in unused model local variables

external_function: modified to pass functions as string names as opposed to function handles to reduce number of calls to mexCallMatlab when use_dll option is invoked

unused model local variables (closes #101)

Windows/64-bit platform

options (closes #98)