One node process for the size-help tooltip tests, not one per render (#898)

[chchatzop]

Fixes #898.

What was wrong

#896 - a text-only change - went red on windows-latest / Python 3.12: TheTooltipInARealEngine.test_a_field_without_a_unit_gets_nothing_added raised subprocess.TimeoutExpired, node not answering within 30 s on a runner that took 333 s for the whole suite. Re-run, it passed (9/9). The test is mine, from #686: the class started a fresh node process for every render - about ten per run - each with its own 30-second timeout, so every render was another chance to lose to a cold start. The three earlier failed CI runs whose logs mention this file were checked: it did not fail in any of them, so this was the first time.

What changed

• setUpClass renders every case the class needs - each size setting in English, MAX_RAR_FOLDER_SIZE in Spanish and French, and one field with no unit - in one node process, with a 120 s timeout (the other real-engine test gives a single node --check 60 s). The tests read their tooltip from that one run.
• What is asserted is unchanged: the same t, fieldHelp and settingsHelpHtml lifted out of app.js, the same dictionaries, the same assertions. The script is still written to a temp file (three dictionaries inlined is past what a Windows command line carries), and the file is removed in a finally.

Verified

• The class runs in 0.15 s locally instead of about ten node starts.
• Mutation: with the unit sentence removed from app.js (settingsHelpHtml), the two tests guarding it still fail; the no-unit test is the control and passes both ways.
• Full suite on current main: 6421 OK (18 skipped).

🤖 Generated with Claude Code

https://claude.ai/code/session_01AP6LSxkr4n9dMFNSNMogmW

[chchatzop]

Superseded by #907 (merged), the same fix for #898. Closing.