Keep translation domains isolated when sharing Babel across apps

[ac1982]

When one Babel instance initializes two Flask apps, Babel.domain_instance caches the first app’s Domain on the shared extension. Requests to the second app can therefore use the first app’s translation catalog and domain, even though the rest of its Babel configuration is application-specific. This is reproducible through ordinary Flask requests rendering Jinja gettext and calling ngettext.

Cache the default Domain on the existing application-owned BabelConfiguration and have Babel.domain_instance delegate to it. This preserves catalog reuse across requests while separating applications and resetting the cache when an app is reinitialized. No global application registry or per-request catalog reload is needed. This addresses a remaining translation-cache path in the multi-app use case from #107.

Five regression cases fail on current master and pass with this change. They use compiled catalogs and cover equal/different domains, both first-request orders, repeated requests and loader call counts, nested contexts, reinitialization, and an explicit request-local Domain override.

Validation on Python 3.8 and 3.12: 35 tests pass, with 3 pre-existing list-translations ordering failures. The unchanged base has those exact same 3 failures (30 passing); #242 separately addresses that ordering. Black 24.4.2 checks the whole repository successfully and the Sphinx documentation builds. Windows/macOS/PyPy were not run locally.

Developed and tested with OpenAI Codex assistance.