Update the Homebrew tap on every version tag

[beetlebugorg]

What changes

The tap job was gated on a tag with no suffix, so a release candidate skipped it. That left the tap update as the one part of the release path a candidate could not prove, which is the opposite of what a candidate is for. The gate is removed.

v1.0.0-rc2  ->  Formula/go-dims.rb
v1.0.0      ->  Formula/go-dims.rb

A candidate now runs the same path the release runs: clone the tap with the token, generate the formula, commit, push.

brew install beetlebugorg/tap/go-dims therefore serves the newest tag, a candidate included, until the next tag replaces it. The installation page says so, and points at dims version for checking which one is installed.

What v1.0.0-rc2 already proved

The candidate ran before it was stopped, and the part that had never executed passed:

success  Build macOS (arm64, macos-latest)
success  Build macOS (amd64, macos-15-intel)
success  Build and Push (linux/amd64)
success  Build and Push (linux/arm64)
success  Push Manifest

So brew install vips, the ANTLR jar wrapper, PKG_CONFIG_PATH, make all on macOS, and the version check all work on real runners for both architectures. Release and the tap job were cancelled with the run.

The tag and its release are deleted, so v1.0.0-rc2 can be cut again. ghcr.io/beetlebugorg/go-dims:1.0.0-rc2 and :next still carry that build; the next tag overwrites them.

Verify

The workflow parses, and the condition reduces to startsWith(github.ref_name, 'v'). The job also keeps needs: [release], so it cannot run unless the release it publishes the formula for succeeded.

The formula content, the sha256 values, and the install were checked when the formula landed: generated against the real v1.0.0-rc1 archives, installed from a local tap under homebrew/brew, dims version reporting 1.0.0-rc1, and brew test passing.