JS: fix the NodeJS::Requireperformance regression

[esbena]

This thouroughly fixes the performance of the regression in #2347.

Caching the charpred of NodeJS::Require seems reasonable since it makes us handle require and import with equal performance in the later stages. When NodeJS::Require is not explicitly cached, we may need to reevaluate non-trivial predicates to identify it, whereas the syntactic import always is identified for free in practice.

I am still running evaluations, but a preliminary evaluation shows that #2347 with cached is faster on definitions.ql than the master branch it was merged into!

[asgerf]

💪

[esbena]

Another successful evaluation: now on top of 1.23: https://git.semmle.com/esben/dist-compare-reports/tree/js/fix-mjs-check_1574947495528.

Do we need a full dist-compare as well, or should we let that be part of the 1.22/1.23 comparison?

[asgerf]

Do we need a full dist-compare as well, or should we let that be part of the 1.22/1.23 comparison?

I'd say let it be part of the 1.22/1.23 comparison as we'd otherwise just delay that comparison.