[kvlonge]

Why

Release Performance fails on every push to main while the current version is tagged but not on PyPI. The job forced that version as the baseline, waited 30 minutes for it to publish, and then exited with a comparison error without measuring anything. The same applied to a release that was tagged but not yet published.

Changes

• The baseline is the newest release PyPI actually serves. A main commit carrying an already released version uses that release once it is published, and the newest earlier release until then.
• A release commit compares with the newest published release before it, never with itself.
• The workflow passes --same-version-baseline for main commits that keep the version instead of forcing the version as the baseline.
• A missing baseline or a failed install still fails the job.
• No DuckShop case: this changes CI tooling, not compiler behaviour.

Verification

• Unit tests for baseline selection: unpublished own release, published own release, release commit after publish, yanked release, first release.
• Against live PyPI, a 0.187.3 candidate now resolves to 0.187.2.
• make check-ci; the workflow step passes bash -n.
• This PR's Release Performance run measures against 0.187.2.