Optimize mesh-data hashing for instancing

[corybarr]

When profiled on Snowden, the Revit addin currently spends 57% of its time hashing mesh data to check for instances. Most of this 57% is in the serialization of the mesh data prior to calculating the hash. This can be optimized.

A simple optimization would be to initially check for the same number of vertices prior to expensive serialization. Other quick tests such as normalized bounding-box ratios might be possible.

It might also be possible to skip mesh-data hashing altogether. The System property of a FamilyInstance looks similar to a shared mesh across instances: