chore(release): bump version to 0.17.0 and rebuild dist

[kalwalt]

Phase A of the release runbook in MAINTAINERS.md. Phase B (promote dev → main, tag, publish) is left to the maintainer.

Why 0.17.0 and not 0.16.1

fast_corners.detect now returns a different, larger corner set for every image (#202). It is a bug fix, but any pipeline tuned against 0.16.0's output will see it shift. A patch number reads as "safe to bump blindly", which is not true here; in a 0.x line the minor is the honest signal.

What's in it

Change	PR
fast_corners.detect per-row candidate buffer off-by-one — detect is now a pure function of its inputs and recovers the corners it was dropping	#203
Bench guards made #202-aware (the bench/ smoke check is not run by npm test)	#203
GitHub Release step made idempotent, so a same-tag retry works	#206
Node pinned to v24.20.0	#204
vite-plugin-dts 5.0.3 → 5.1.0	#199

Steps performed

1. package.json → 0.17.0
2. Lockfile regenerated with npm 11 (npx npm@11 install) — lockfile root and packages[""] both confirmed at 0.17.0
3. npm run build-ts → dist/jsfeatNext.js, dist/jsfeatNext.mjs, types/ rebuilt and committed
4. npx git-cliff --unreleased --tag 0.17.0 --prepend CHANGELOG.md

Verification

• npm test — 356 passed, 28 files
• npm run bench:smoke — exit 0
• npm pack --dry-run — 43 files, 88.7 kB packed / 295.1 kB unpacked, version 0.17.0
• prettier --check — clean
• The built artifact was exercised directly, not just the source: loading dist/jsfeatNext.mjs and running detect against a pool poisoned with three different values returns an identical 33-corner set each time, confirming the fast_corners.detect reads uninitialised memory and drops each row's last corner #202 fix is present in the shipped bundle and that detect is pool-independent there.

Note on the changelog entry

CHANGELOG.md had a hand-written ## Unreleased section describing #202 in full prose. It was removed before running git-cliff so the entry would not be duplicated, and the generated section now follows the repo's usual git-cliff format (commit subject + hash) like every other release. The long-form explanation of the defect lives in docs/implementation-notes.md and in #202.

🤖 Generated with Claude Code

[qodo-code-review]

ⓘ Qodo reviews are paused because your trial has ended. Ask your workspace admin to add credits to resume reviews. Manage billing