Docs: say when to turn on the JVM AOT cache (opt-in, with measured figures) #247

The JVM package (arcadedb-jvm) has an opt-in JDK 25 AOT cache (ARCADEDB_AOT=1 or start_jvm(aot=True), src/arcadedb/aot.py, merged in #31), but no page of the user docs says what it does, when it helps, or what it costs. Only the docstrings of start_jvm mention it.

A timed A/B on one laptop (JDK 25, arcadedb-jvm 26.11.1.dev0, ABBA order, 19 rounds per arm, 18 clean pairs per figure, with an A/A control) decided whether the cache should be the default. It should not: the first script ran faster with the cache (ratio 0.653, 95% interval [0.648, 0.658], first import to first openCypher result), but the steady-state figures showed no gain I can rely on, and the 300,000-row insert_many interval ended at 1.036, above the 1.03 no-harm bar written before the run. So the cache stays opt-in, and the study's rule says to document when to enable it.

The docs should say: what the cache does, how to turn it on and off, when to turn it on (short-lived processes where start-up dominates) and when not to bother (long-running services), the one-time costs (a training run and a cache file on disk), and the measured figures with their intervals, stated as one study on one machine.

I will open a docs-only PR (docs/modules/ROOT/pages/installation.adoc, plus a link from which-package.adoc).