deprecated functions

[Jason Riedy]

And John W. Eaton writes:
> But exactly what would be catastrophically bad about it?

Three differences pop out immediately, in decreasing order of
anyone likely caring:

  - There are no single-precision sparse matrices, so single-prec
    diagonal and permutation matrices need to devolve into dense.

  - eye(1) * sparse (1, 1, 1) would produce a 1x1 sparse matrix rather
    than a scalar.  Likely some odd problems with 0-dim matrices, too.

  - Any code relying on the exceptional behavior will be broken. That
    is, (diag * sparse) / 0 will not have NaNs filling the zeros of the
    sparse matrix.

Obviously, I won't consider those catastrophic, but they are
differences.

Jason