Remote Config v2: guard reads before activation

[shameondev]

Summary

• preload the exact identity-scoped snapshot off the read path and prepare a crash-safe first activation before SDK readiness
• assert actionable misuse in debug; in release, allow one SDK-lifetime implicit activation on the first read, then require explicit activation
• keep current synchronous and free of storage calls, preserve immutable held snapshots, and deliver updates outside SDK locks
• distinguish missing, corrupt, failed, and persistence-failed preload state and harden bounded legacy-v1 validation

Concurrency guardrails

Tests pin stale A→B→A and same-generation preload races, token binding, concurrent first reads, persistence recovery, observer reentrancy/lost wakeups, lifetime behavior across identities, and post-preload fetch admission. Stale work cannot overwrite a newer durable candidate or consume the lifetime activation opportunity.

Verification

• ./gradlew :sdk:test detektAll :sdk:assemble --no-daemon — BUILD SUCCESSFUL
• 82 focused Remote Config tests, including 22 read-guard adversarial tests
• independent multi-pass review — CLEAN
• git diff --check

Boundary

This is an internal lifecycle foundation stacked on #862. It does not add a public Remote Config API, URL, or production enablement.

[shameondev]

Консолидировано в release-train PR (один PR на репозиторий). Ветка сохранена.