FIX support axis=-1 and -2 for sparse reduce methods

[jnothman]

Fixes #3249

[pv]

A better idiom could be to add to the top-level axis-using functions

if axis < 0:
    axis += self.ndim

and then proceed as usual.

[pv]

Please also fix getnnz while you're at it ;)

[jnothman]

A better idiom ...

If you rather.

[cowlicks]

I was unaware of the axis=-1 feature when I implemented this! Thanks for the good work @jnothman !

[jnothman]

@cowlicks, you can blame @jaimefrio for demanding it...

[coveralls]

Coverage remained the same when pulling 5936e4c on jnothman:neg_axis into c683ff9 on scipy:master.

[pv]

Thanks, merged.