Remove redundant startup reloads and quiet routine lifecycle logs

[remdui]

Summary

Reduce startup work and console noise for every FeatureFramework host, including both ServerFeatures and ProxyFeatures.

Root cause

Startup storage preflight created a temporary feature/context and called the full prepare(...) lifecycle. Managed preparation injects defaults and reloads config/localization. Actual feature activation then creates the real feature/context and runs the same preparation again.

That meant routine boot performed two config/localization reload phases per enabled feature and emitted duplicate Localization files reloaded... INFO messages.

Fix

• add a dedicated prepareStorage(...) preflight boundary;
• managed preflight now only materializes config/message defaults;
• actual activation performs the one runtime config/localization reload;
• startup localization reloads use a quiet DEBUG path;
• explicit/operator-triggered localization reloads continue to log at INFO;
• add DEBUG support to the shared framework logger with JUL/FINE and SLF4J/DEBUG mappings;
• move routine feature-scoped DataProvider scope/connection lifecycle messages to DEBUG;
• preserve backward compatibility for custom FeatureHostContext, FeatureLocalization, and FrameworkLogger implementations through default methods.

Architecture

This deliberately does not merge feature-scoped ORM contexts. Their lifecycle/entity isolation is intentional; startup detail should be quieter without weakening resource ownership boundaries.

The DataProvider ORM/entity/reference-count logging and explicit non-JTA Hibernate bootstrap are handled in the companion DataProvider PR.

Tests

Added regression coverage proving:

• storage preflight injects config/message defaults without runtime config/localization reloads;
• real runtime preparation reloads config and localization exactly once;
• startup uses the quiet localization path while explicit reload remains a separate operator-facing API.