perf(global): memoise pluralize()/singularize() results #4150

Measured on: RustCFML v0.693.0 (the CI pin), develop 0255f0d53, production mode, Apple M3 Ultra.

Problem

pluralize() / singularize() ($singularizeOrPluralize in vendor/wheels/global/strings.cfm) redo all their setup on every call:

• rebuild the irregulars list from settings;
• clean the uncountables list;
• rebuild a 2D array of about 20 regex rules from a comma list;
• try each rule with ReFindNoCase.

Results depend only on (text, which) and the inflection settings. The framework itself calls these on a small, fixed vocabulary of model, table and route names.

Measured on RustCFML, per call, median of 3 runs:

On a 50-row list page that calls pluralize once per row, the CFML sampling profiler put pluralize at 18.5% of request samples. That share comes from my test view; the per-call cost is the general number.

Not measured yet: Lucee, Adobe, BoxLang.

Proposal

• Memoise results in an application-scoped cache keyed by which|text. Cache the uninflected result, and apply count / returnCount formatting after the lookup.
• Clear the cache whenever uncountables or irregulars change via set(), and on reload.
• Optionally build the rule array once instead of per call.
• Bound the cache size (or cache only when the text is short) so user-supplied text can't grow it without limit.

Acceptance

• Existing inflection specs pass unchanged on all engines.
• New specs: a cache hit returns the same value; changing irregulars / uncountables invalidates the cache; the cache stays bounded.
• Before/after per-call timing posted on the PR.