How to handle singularities in the correction matrix?

[efaulhaber]

We didn't have this problem with TLSPH, but when using the correction matrix for WCSPH, for example for kernel gradient correction or density diffusion, it might become singular and can't be inverted.

How do we correctly handle this case? Set the inverse to zero? Or the identity? I chose the latter in #272.

[svchb]

This is also related to #271 and #270. As well as adaptive support.

For now one can use the identity matrix. Setting the inverse to 0 is plainly wrong except for cases in which the matrix does not matter.

[svchb]

In #261 it currently exits with an error.

[svchb]

To merge #261 and #272 we should use similar conditions i.e. set the identity matrix for a critical condition number.