Need multi-attribute query stacking for dense arrays

[ihnorton]

The current API requires manually stacking the returned OrderedDict.values() in order to treat the result as stacked array (assuming the types are homogenous; this can be useful in some cases). This is inefficient due to double memory usage, copying, etc. We should add an API to return a multi-attribute query as a stacked N+1 array (can pre-allocate the entire buffer and read into it for each attribute-level query).