feat: add a dry-run input to the GitLab CI component

[lesnik512]

Closes #60.

The GitLab component now matches the GitHub Action: a dry-run input threads --dry-run to semvertag tag.

• Input shape. dry-run is a GitLab type: boolean input with default: false, unlike the GitHub Action's string 'false'. GitLab renders a boolean input as the bare word true/false, which are shell builtins, so the script uses $(if $[[ inputs.dry-run ]]; then echo --dry-run; fi) on the same line as uvx. That keeps the invocation at script[0], where the gate already looks for the version pin. A boolean also lets GitLab reject anything other than true/false when the pipeline is created, instead of the script quietly treating a typo as "off".
• Version floor. The component's semvertag floor moves from >=0.1 to >=0.5.0, the first release with --dry-run, matching action.yml.
• Descriptor gate. The gate now requires spec.inputs to be exactly {strategy, dry-run}, dry-run to be a boolean defaulting to false, script[0] to wire the input onto --dry-run, and the semvertag floor to be at least 0.5.0. Each rule has a negative fixture.
• Docs. The GitLab guide gets a "Dry run" section for the inline job and the component input.

Not done here: the quick-start snippet in docs/providers/gitlab.md still pins semvertag>=0.1,<1. It never passes --dry-run, so it still works, and the new section tells readers to raise the floor when they add the flag.

Verified with just lint-ci, just test-ci (480 passed, 100% coverage), just adr-check and just docs-build. The script line was checked in a local POSIX sh (it expands to tag --dry-run for true and tag for false). It has not been run in a real GitLab pipeline; the component isn't published to the Catalog yet.