handle inf and nan in features

[cavalab]

handle inf and nan in features

[lacava]

inf: output max system size for stack type
nan: output 0

• make safe function in base Node class
• call safe on return values of evaluate function for division, exponent, and log nodes

[lacava]

reopening this issue because there are still issues with inf and nan features, especially when normalizing features. need to come up with a better strategy to address this.

[lacava]

it seems like nans were coming from eigen's normalize function. this block manually checks for near-zero columns in X before normalizing.