Release v1.4.0

Changes

• BREAKING CHANGE: Added last_published_index and header magic in shared memory header
  • Fixed shared-memory attach race by adding an init-state handshake with readiness wait and legacy fallback
  • Added last-published tracking so read_last() only returns published data; added a regression test
  • Changed read_last() return signature to std::pair<T*, uint32_t>
• Updated read_last() documentation for legacy shared-memory segments
• Relaxed memory ordering in the shared atomic cursor path for lower overhead
• Fixed release workflow changelog extraction to match version headers with trailing dates