v0.11.0

release: 0.11.0

Bumps the two version files and writes the changelog. Both matter for different
reasons, and one of them would have failed silently.

npm rewrites package.json from the tag at publish time, so the stale 0.10.2 there
was cosmetic — but a second place the version can be wrong is a second place it
can be read wrong, so it is bumped to agree.

PYPI DOES NOT. `publish-pypi` runs `uv build` against the tagged commit and takes
the version straight from pyproject.toml, with `skip-existing: true` on the upload.
So tagging without this bump would have built 0.10.3, matched the release already
on PyPI, skipped, and reported success. npm at 0.11.0, PyPI still at 0.10.3, and
nothing anywhere saying so.

The changelog leads with the two bugs that existed ONLY in the artifact people
install — `gecko ingest-gate` crashing in every published binary since the check
was written, and `[ocr]` never once surviving a release build since 2026-08-08 —
because that is the shape of what this release closes, and both were invisible to
a green suite.

Five weeks and 240 commits behind the tag that should have carried them.

Full suite 4,972 passed / 61 skipped / 3 xfailed, mypy clean on 250 files.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J7EsRmt4kG9xVx32EyGjKz