Csc-pysparse

--------------------------------------------------------------------------------
Moved modules inside "pysparse" name space.

Updated to use the new numpy, but still compatible with the old Numeric.

---------------------------------------------------------------------------------
Fixed a memory leak in ll_mat.keys(). (Reported by c chui)

Improved the performance of ll_mat.scale(), ll_mat.values() and ll_mat.items().

---------------------------------------------------------------------------------
Modules using the pysparse C API can now define NO_IMPORT_SPMATRIX
together with SPMATRIX_UNIQUE_SYMBOL to avoid multiple definition of
the SpMatrix_API pointer. This fix was suggested by Erica Madeleine Ligner.

---------------------------------------------------------------------------------
Applied patches 1089844, 1089837, 1089831 (supplied by Daniel Wheeler
& Jonathan Guyer): Added three new ll_mat methods:
update_add_at(b,id1,id2), put(b,id1,id2), take(b,id1,id2), fixed OS X
compatibility issues, Numpy compatibility fixes.

Applied some additional fixes for 64-bit machines.

PySparse now works with Numeric 23.7 (also on 64-bit machines).

[1111727] Replaced include <malloc.h> by include <stdlib.h> to fix
compatibility issue on OS X (reported by Jonathan Guyer).

Removed reference to libm.a which broke compilation on Win32 (reported
by Mark Dausch).

---------------------------------------------------------------------------------
added new umfpack module. (by Nathaniel C. Domingo)

added Umfpack library source files to distribution (by Nathaniel C. Domingo)

added new module sparray, implementing a multidimensional sparse array
(by Pascal Barbedor)

fixed bug in SSOR preconditioner (precon module) for omega != 1.0 and
steps > 1

---------------------------------------------------------------------------------
added Xavier Vasseur's gmres solver to the itsolvers module. Thanks Xavier!