chore(release): cut v1.0.0 — a new foundation for production Vision AI

[stwilt]

📋 What does this PR do?

OpenFilter 1.0. Bumps VERSION to v1.0.0 and adds the 1.0 release notes to RELEASE.md. v0.2.1's entry stays below as preserved history of the preview cycle.

🔍 Why is this needed?

OpenFilter 1.0 transitions the runtime from an evolving framework into a stable, production-grade platform. The typed declarative-config surface (FilterConfigBase, FilterOutputSchema, the openfilter emit-schema CLI) introduced during the v0.2.x preview cycle is committed as 1.0's public API. Filters can rely on these surfaces under SemVer.

OpenFilter 1.0 ships 2026-05-18.

🧪 How was it tested?

• VERSION and RELEASE.md agree on v1.0.0.
• changelog-parser-action parses the new header cleanly.

🔗 Related Issues

• FILTER-441 — typed FilterConfigBase + emit_schema
• FILTER-442 — emit-schema CLI
• FILTER-444 — FilterOutputSchema + shape catalog
• FILTER-452 — $id inheritance fix
• FILTER-461 — Python 3.10 shutdown race
• FILTER-462 — release-pipeline cascade trigger

✅ Checklist

• I have read and agreed to the terms of the LICENSE
• I have read the CONTRIBUTING guide
• I have followed the coding style
• I have signed all commits in compliance with the DCO (git commit -s)
• I have added or updated tests as needed (no test impact; release-notes + version-bump only)
• I have added or updated documentation as needed (this PR is the documentation change — release notes for 1.0)

[shingonoide]

Thanks for cutting v1.0.0. The VERSION / RELEASE.md pairing is consistent and the version-match check at .github/workflows/create-release.yaml:67-80 will pass. Three observations before workflow_dispatch fires.

1. Release date stamped one day in the future — RELEASE.md:5

## v1.0.0 - 2026-05-18

Today is 2026-05-17. If the workflow is triggered today the published GH release notes will carry tomorrow's date while the actual PyPI/Docker publish timestamps will be today's. Either advance the heading to 2026-05-17 or hold the workflow_dispatch until 2026-05-18.

2. ResolveHint is a public export not covered by stability commitments — openfilter/filter_runtime/__init__.py:8,56; RELEASE.md:39-43

ResolveHint is imported and listed in __all__ of openfilter.filter_runtime, but the "Stability commitments" section commits only FilterConfigBase (with Managed/Resolve), FilterOutputSchema+shape catalog, and openfilter emit-schema. ResolveHint is the Literal that parameterizes the resolve= argument of Managed/Resolve, so callers annotating their config fields with it are implicitly relying on it. Either add it to the committed surface or document it as explicitly excluded so a future rename does not require a 2.0 bump.

3. Conflicting `dev` groups across PEP 621 and PEP 735 — `pyproject.toml:61-71` and `:123-127`

`dev` is defined under `[project.optional-dependencies]` (PEP 621, ~10 packages including `pytest`, `twine`, `build`) and again under `[dependency-groups]` (PEP 735, only `pytest>=9.0.2` and `pytest-benchmark>=5.1.0`). The two groups give silently different environments depending on the tool: `pip install ".[dev]"` resolves the PEP 621 set while `uv sync --group dev` resolves the PEP 735 set. For a v1.0.0 cut it would be worth either consolidating to one source of truth or renaming one of the groups (e.g. `dev-bench`) so the divergence is explicit.

No blockers found. Items unchanged from prior reviews still apply: VERSION match check, changelog-parser format, breaking-change docs for `OTLP_GRPC_ENDPOINT_SECURITY` and `frame.image` read-only, cascade tag pattern matching `v1.0.0`, and `GH_BOT_USER_PAT` wiring all look correct.

[lucasmundim]

Code review

Approving. Two non-blocking suggestions inline:

1. v1.0.0's preamble frames the release as additive ("filters built against earlier versions keep running unchanged"), and ### Breaking Changes only lists the OTLP scheme inference (#90). The Frame.image read-only break introduced in v0.1.28 is a real source-incompatible change with explicit migration notes. Consumers who skip the preview cycle and jump straight to 1.0 lose that signal. Either re-surface it under v1.0.0 ### Breaking Changes, or soften the preamble.

2. Section order in the new v1.0.0 entry is Fixed -> Removed -> Infrastructure. Keep-a-Changelog and the prior v0.1.30 entry put Removed before Fixed. Same nit raised on #96 but not addressed; recurring here.

[shingonoide]

Thanks. Re-review against ff1566e:

1. Release date — RELEASE.md:5 now reads ## v1.0.0 - 2026-05-17. Addressed.
2. ResolveHint in stability commitments — RELEASE.md:41 explicitly names the ResolveHint literal alongside FilterConfigBase / Managed / Resolve. Addressed.
3. PEP 735 [dependency-groups] — removed from pyproject.toml; PEP 621 [project.optional-dependencies].dev remains the single source of truth, consistent with the .[all,dev] install path at Makefile:82. Addressed.

Approving.

[lucasmundim]

Re-review against bfba5af. Both items from my prior approval are addressed:

1. frame.image read-only break is re-surfaced under v1.0.0 ### Breaking Changes with explicit migration guidance and the ValueError symptom; preamble at RELEASE.md:7 softened to scope behavioral changes to the list. Addressed.
2. Section order swapped to Removed -> Fixed -> Infrastructure, matching Keep-a-Changelog and v0.1.30. Addressed.

Approving. One small date-stamp observation inline.

[lucasmundim]

Re-review against c31ede6. Date bump from 2026-05-17 → 2026-05-18 addresses my prior inline comment — RELEASE.md:5 now matches today's UTC date and the imminent publish timestamps. No other changes since bfba5af. VERSION/RELEASE.md heading agree on v1.0.0 and the create-release version-match check will pass.

Approving. Safe to fire workflow_dispatch.