feat: add EverOS memory integration

[Yangtze-Seventh]

Closes #3875.
Related to deepset-ai/haystack#12499.

Summary

• Add everos-haystack, a Cloud-backed long-term memory integration using the canonical EverOS /api/v2 contract.
• Add EverOSMemoryStore with add, flush, search, typed result conversion, owner/scope isolation, Haystack filter conversion, serialization, and structured API errors.
• Add Writer/Retriever components and Agent-facing remember/recall tools with IDs injected from Haystack Agent State.
• Add runnable component and Agent examples plus documentation covering EverOS's dual user/agent memory model and eventual index consistency.
• Register the integration in the repository inventory, labeler, coverage workflow, and nightly CI matrix.

EverOS-specific behavior

EverOS extracts durable memory from message streams rather than treating every message as one memory. flush_on_write exposes an explicit extraction boundary for short jobs, while the default behavior allows normal semantic boundaries.

User episodes/profiles and agent cases/skills remain separate first-class tracks. Searches require exactly one of user_id or agent_id; app_id, project_id, and optional session_id preserve EverOS scope isolation. Returned memories are Haystack system ChatMessage objects with their EverOS type, source IDs, scope, scores, and request_id preserved in metadata.

Test plan

• hatch run fmt-check
• hatch run test:types
• hatch run test:unit — 50 passed
• hatch run test:unit-cov-retry — 95% coverage
• Lowest direct dependencies (haystack-ai==2.19.0, httpx==0.27.0) — 50 passed
• hatch run docs
• hatch build
• Live EverOS Cloud add/flush/search tests — 2 passed, including exact recall and wrong-user isolation

Maintainer setup

The live tests skip when EVEROS_CLOUD_API_KEY is unavailable. To enable the nightly Cloud contract tests after merge, a repository secret with that name will need to be configured. No credential is included in this PR.

[CLAassistant]

All committers have signed the CLA.

[github-actions]

Hi @Yangtze-Seventh, thanks a lot for contributing a new integration! 🙏

This PR looks like it adds a new integration (everos), but we couldn't
find a linked issue labeled new integration.

Every integration we merge is maintained, tested nightly, and released by us, so before reviewing the code
we'd like to understand the adoption of the technology you're integrating with. Could you please
open a New Integration Proposal
and fill it out? The Adoption signals section (GitHub stars, PyPI downloads over the last 30 days,
release activity, community demand) is what helps us decide whether to add the integration to Haystack.

Then reference the issue in this PR's description (for example, by adding Related to #<issue-number>)
and this comment will disappear automatically.