TODO

Functionality
=============
- Add support for BLIS and fused Trmv's
- Clean up pivoted LU and pivot applications (especially in sequential cases)
- Add transpose option to LeastSquares
- Pivoted QR and LQ decompositions 
- Lower version of reduction to bidiagonal form
- Reduction to Hessenberg form
- Compressed pseudoinverse solves which avoid unnecessary backtransformations
- Implement sequential versions of Trr2k and HermitianEig
- Add wrappers for more LAPACK eigensolvers
- More explicit expansions of packed Householder reflectors
- More TwoSidedTrmm/TwoSidedTrsm routines
- Nonsymmetric (generalized) eigensolver
- Add means of detaching and reattaching the 'viewing' processes for a 
  DistMatrix so that collective routines like 'Get' can still be called

Maintenance
===========
- Separate out local blas-like and lapack-like operations
- Move most functions out of 'elem::internal' namespace and into a namespace
  corresponding to its particular class of algorithm, e.g., elem::svd
- Further improvements for the speed of the build process
- Consistent handling of unblocked routines
- Standardize on impl/decl pairs
- Switch to C++11 for better 64-bit integer and std::complex support, as well as
  std::thread, extern templates, and syntactic sugar