docs: add release notes for version 0.7.0 (#5408)

## Summary

This PR adds release notes for the upcoming 0.7.0 release.

### Files

- **`RELEASE_NOTES.md`** — Updated top-level summary with 0.7.0
highlights and breaking changes table
- **`docs/releases/RELEASE_NOTES_0.7.md`** (new) — Comprehensive 0.7.0
release notes with migration instructions, before/after code examples,
contributor credits, and upgrade guide

### What's Covered in 0.7.0

- 13 hard breaking changes with migration instructions
- 3 behavior changes
- Reasoning output support in Responses API
- Comprehensive observability metrics (API, inference, vector IO)
- Structured logging via structlog
- Inline neural rerank for RAG
- Inline Docling provider for PDF parsing
- Infinispan vector-io provider
- Background response cancellation
- Connector API promoted to v1beta
- FastAPI router migration complete (@webmethod removed)
- 4 performance improvements (lazy-loading)
- 3 CVE security fixes
- 20+ bug fixes
- 12+ documentation updates
- 15+ CI/CD improvements

## Test Plan

- [x] Markdown renders correctly
- [x] markdownlint passes
- [x] Pre-commit whitespace checks pass
- [x] All PR links use correct `llamastack/llama-stack` repo URLs

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Signed-off-by: Artemy <ahladenk@redhat.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Sébastien Han <seb@redhat.com>

Author: 3 people

RELEASE_NOTES.md

@@ -4,6 +4,44 @@ This file provides a summary of each release. For detailed release notes with mi
 
 <!-- New releases go here, at the top. See docs/releases/ for detailed notes. -->
 
+## [0.7.0](docs/releases/RELEASE_NOTES_0.7.md) - 2026-04-01
+
+Release 0.7 is a major release focused on completing the transition to OpenAI API conformance, introducing comprehensive observability metrics, and significant API cleanup. This release removes the fine-tuning API, completes the FastAPI router migration, removes legacy providers (TGI, HuggingFace), renames core concepts for clarity, and adds structured logging via structlog.
+
+### Highlights
+
+- **Agents API renamed to Responses API** aligning with OpenAI naming ([#5195](https://github.com/llamastack/llama-stack/pull/5195))
+- **Reasoning output support** in Responses API ([#5206](https://github.com/llamastack/llama-stack/pull/5206))
+- **Comprehensive observability metrics** for API, inference, and vector IO ([#5201](https://github.com/llamastack/llama-stack/pull/5201), [#5320](https://github.com/llamastack/llama-stack/pull/5320), [#5096](https://github.com/llamastack/llama-stack/pull/5096))
+- **Structured logging via structlog** with key-value output ([#5215](https://github.com/llamastack/llama-stack/pull/5215))
+- **Inline neural rerank for RAG** without external services ([#4877](https://github.com/llamastack/llama-stack/pull/4877))
+- **Inline Docling provider** for structure-aware PDF parsing ([#5049](https://github.com/llamastack/llama-stack/pull/5049))
+- **Infinispan vector-io provider** for distributed vector storage ([#4839](https://github.com/llamastack/llama-stack/pull/4839))
+- **Connector API promoted to v1beta** ([#5129](https://github.com/llamastack/llama-stack/pull/5129))
+- **FastAPI router migration complete** with `@webmethod` removal ([#5248](https://github.com/llamastack/llama-stack/pull/5248))
+- **Performance**: lazy-loading of torch, numpy, faiss, and braintrust to reduce startup memory ([#5116](https://github.com/llamastack/llama-stack/pull/5116), [#5118](https://github.com/llamastack/llama-stack/pull/5118), [#5078](https://github.com/llamastack/llama-stack/pull/5078))
+
+### Breaking Changes
+
+| Change | Type | PR |
+|--------|------|-----|
+| Fine-tuning API removed | Hard | [#5104](https://github.com/llamastack/llama-stack/pull/5104) |
+| `meta-reference` providers renamed to `builtin` | Hard | [#5131](https://github.com/llamastack/llama-stack/pull/5131) |
+| `knowledge_search` renamed to `file_search` | Hard | [#5186](https://github.com/llamastack/llama-stack/pull/5186) |
+| Agents API renamed to Responses API | Hard | [#5195](https://github.com/llamastack/llama-stack/pull/5195) |
+| `tool_groups` removed from public API | Hard | [#4997](https://github.com/llamastack/llama-stack/pull/4997) |
+| TGI and HuggingFace providers removed | Hard | [#5333](https://github.com/llamastack/llama-stack/pull/5333) |
+| `register`/`unregister` model endpoints removed | Hard | [#5341](https://github.com/llamastack/llama-stack/pull/5341) |
+| `@webmethod` decorator removed | Hard | [#5248](https://github.com/llamastack/llama-stack/pull/5248) |
+| `rag-runtime` provider renamed to `file-search` | Hard | [#5187](https://github.com/llamastack/llama-stack/pull/5187) |
+| Duplicate `dataset_id` parameter removed | Hard | [#4849](https://github.com/llamastack/llama-stack/pull/4849) |
+| `/files/{file_id}` GET response unified | Hard | [#5154](https://github.com/llamastack/llama-stack/pull/5154) |
+| OpenAI API schema transforms | Hard | [#5166](https://github.com/llamastack/llama-stack/pull/5166) |
+| `starter-gpu` distribution removed | Hard | [#5279](https://github.com/llamastack/llama-stack/pull/5279) |
+| `sentence_transformers` `trust_remote_code` defaults to `False` | Behavior | [#4602](https://github.com/llamastack/llama-stack/pull/4602) |
+
+See the [full release notes](docs/releases/RELEASE_NOTES_0.7.md) for migration instructions and detailed upgrade guide.
+
 ## [0.5.0](docs/releases/RELEASE_NOTES_0.5.md) - 2026-02-05
 
 Release 0.5 brings significant improvements to API consistency, OpenAI conformance, provider capabilities, and a major architectural refactoring of all APIs to use FastAPI routers.