optimisation: factor out common prefixes in large alternatives

[lsf37]

In #715 one of the features of that particular spec is a very large "or" of strings.

My hypothesis there was that this tends to blow up the initial DFA construction, and that factoring out any common prefixes automatically could potentially reduce that blow-up quite a bit.

This issue is for

• figuring out if that hypothesis is true
• if the optimisation would be worth the additional processing time
• if yes, implement it