Improve performance of repeated XPath queries over secondary instances

[lognaturel]

Given a form that has a secondary instance
And multiple XPath queries against that secondary instance with a common subexpression
Then queries after the first should be faster than the first

select-from-file-50k.xlsx with external secondary instance external.csv (GSheets source) is big enough to take noticeable time on most devices.

There could be limitations to what kinds of expressions can benefit from the intervention (e.g. only single equality predicates such as name=${selected})