Fix stale refresh generation handling after PR #386

[karthiknadig]

Summary

The original refresh deduplication work in #383 left a follow-up correctness gap after PR #386 was merged.

A refresh that started under an older configuration generation could still:

• emit stale environment, manager, or telemetry notifications after a newer configure, and
• reserve the one-time missing-environment follow-up work in a way that could block the current generation from reporting it.

Expected behavior

Once configuration generation advances, stale refresh work should not:

• sync locator refresh state back into shared locators,
• emit refresh notifications for the old generation, or
• permanently steal the missing-env reporting slot from the current generation.

Proposed fix

• make refresh notification emission generation-aware and atomic with the generation check,
• keep stale refresh state sync blocked,
• replace the missing-env one-shot boolean with generation-stamped reservation semantics so newer generations can supersede stale reservations,
• add regression tests for the stale-generation reporter and missing-env reservation behavior.

Context

This follow-up comes from Copilot review comments discovered after the original PR was merged.