chore: wire up force_publish input in release workflow

[Rodriguespn]

Summary

• Adds workflow_dispatch trigger with a force_publish boolean input so release artifacts can be re-published manually without a new release-please run
• Derives the release tag from the latest git tag when release-please did not create one (e.g. when re-running after a failed upload)
• Adds --clobber to gh release upload to overwrite existing assets instead of erroring

[gregnr]

Thanks @Rodriguespn - two thoughts:

1. The failure mode for agent-skills and supabase-plugin is a lot different than MCP - MCP is publishing to NPM, which inherits the normal coordination challenges which makes a manual force_publish useful in the event that something goes wrong. These repos only upload assets to a GitHub release. Part of me is wondering if we're over engineering a use case we'll never have?
2. If we do still want these, my other concern is about accidentally uploading the wrong assets - see this comment on the MCP repo.

[Rodriguespn]

Thanks @Rodriguespn - two thoughts:

1. The failure mode for agent-skills and supabase-plugin is a lot different than MCP - MCP is publishing to NPM, which inherits the normal coordination challenges which makes a manual force_publish useful in the event that something goes wrong. These repos only upload assets to a GitHub release. Part of me is wondering if we're over engineering a use case we'll never have?
2. If we do still want these, my other concern is about accidentally uploading the wrong assets - see this comment on the MCP repo.

Thanks for the feedback @gregnr. We recently had this failed release when we created the Changelogs. The commit was a chore so it didn't trigger a version bump on the root directory but since it was the first time the changelog was being generated, a set of trailing commits with feat and fix triggered the version bump for each skill. My understanding is that this is a one time thing and future commits with feat: and fix: will always trigger the root version.

Imo reinforces that there should be an agent skill versioning standard so clients can rollback on previous/stable skills versions. Agent skills are different than npm packages as they (usually) don't have code but a change in the tone, phrasing or set of instructions can change the way agents behavior when reading the skills. This is just something that I have been thinking about if we should adopt our own agent skills versioning system. Not an action item, just a thought

To summarize, this might be over engineering for our agent skills and plugin use cases. Closing this and supabase-community/supabase-plugin#39