Fill the whole version gap when bumping wgc

[gausie]

What

bump-wgc.yml previously bumped the formula to exactly the dispatched target version. When several wgc versions are released close together, every version between the tap's current formula version and the target got skipped — no formula commit, no bottle.

This makes the tap derive the gap itself and build/publish the whole range, oldest-first, each with its own bottle.

How

• resolve — reads CUR from Formula/wgc.rb, resolves T (client_payload.version || inputs.version || npm view wgc version), and computes the ordered MISSING list of all stable x.y.z versions strictly > CUR and <= T via semver sort -V. Computes url + sha256 per version once. Emits changed, a meta JSON map, and a build matrix (MISSING × [macos-14, ubuntu-latest]).
• bottle — driven by fromJSON(needs.resolve.outputs.matrix), parameterised per matrix entry. Ensures a per-version wgc-<version> release exists, uploads that version's bottle tarball, and uploads bottle JSON as bottle-json-<version>-<os>.
• publish — downloads all bottle JSON artifacts, then loops over MISSING ascending: perl-edits the formula to each version's url/sha, merges that version's bottle JSONs, and commits wgc <version>. Pushes once at the end.

Concurrency group (bump-wgc, cancel-in-progress: false), the workflow_dispatch version input, and the npm-tarball / brew bottle mechanics are unchanged. The 24h cooldown stays enforced cosmo-side, so any version <= T is safe to build.

Testing

Dry-ran the resolve/publish shell against the live npm version list (no brew/releases/push):

• Rolled-back CUR=0.122.0, target 0.125.0 → MISSING = 0.123.0, 0.124.0, 0.125.0; matrix = 6 entries; publish order ascending.
• No-op case (CUR == T) → changed=false.
• Single-version gap → just the target.
• YAML parses.