Test and improve performance #19
Description
Good performance is one of the design goals of this STAM implementation. I
consider this to include both efficient run-time execution (CPU time), as well as resource
consumption (memory), and we often find a trade-off between the two.
The library implements a fair amount of benchmarks (run cargo bench) to quantify this.
Most insightful, however, are comparisons with other systems. The comparison with TextFabric is most
notable and most informative in this, especially with regard to searching/querying.
Performance is constrained by the way the STAM model itself is designed. Its
aim to be flexible in supporting multiple annotation paradigms, and possibly
act as a pivot model between multiple, implies the implementation can not
always be optimised as much as others. I'm again comparing with TextFabric here
where for instance certain queries may map more directly onto internal
structures.
Further tests and possible refactoring rounds are needed to ensure performance is the best we can do.