Report the install directory in skills get output

[Yusufihsangorgel]

Fixes #552.

After installing, the per-agent summary named the agent but not where the
skills actually went. It now names the directory:

Before:

Installed 2 skill(s) for generic.
Installed 2 skill(s) for claude.

After:

Installed 2 skill(s) for generic at .agents/skills.
Installed 2 skill(s) for claude at .claude/skills.

This follows the display you added to skills list in #554 (d1405de), whose
header is claude (installed at .claude/skills): — same project-relative
directory, same at preposition. The value here is agent.skillsRelativePath
rather than p.split(p.relative(adapter.skillsDirectory, from: rootPath)).join('/');
I checked all six adapters on multiple roots and the two produce identical
strings, so this avoids pulling package:path into get_skills.dart while
staying consistent with what list prints.

Not included: a location on each per-skill line. The directory is constant per
agent, so repeating it on every skill is noise, and InstalledSkillInfo
carries no path — threading one through _installSkills →
_PackageInstallResult → SkillInstallResult would multiply the diff for that
noise. Happy to do it if you disagree.

This touches get_skills.dart and the ## 1.0.0-wip CHANGELOG, the same two
files as your open #568. I merged the two locally and they apply cleanly, so
there's nothing to coordinate.

Test covers two agents with different directories in a single get run.

[gemini-code-assist]

Code Review

This pull request updates the get command's summary message to display the actual installation directory for each agent, and adds a corresponding test to verify this behavior. The reviewer suggested using addTearDown to cancel the logger subscription in the new test to prevent potential resource leaks if the test fails.

[gemini-code-assist]

[CONCERN] If the test fails or throws an exception before reaching await subscription.cancel();, the logger subscription will leak and continue to listen to log events during subsequent tests. This can cause test flakiness or side effects.

Consider using addTearDown to guarantee that the subscription is cancelled when the test finishes.

      final logMessages = <String>[];
      final subscription = Logger('skills get').onRecord.listen((r) {
        logMessages.add(r.message);
      });
      addTearDown(subscription.cancel);

      final getCommand = GetCommand(
        dialogSupport: null,
        gitRunner: GitRunner(isAvailableOverride: () async => false),
      );
      final runner = SkillsCommandRunner('skills', 'Test')
        ..addCommand(getCommand);

      await runner.run([
        'get',
        '--directory',
        projectPath,
        '--agent',
        Agent.generic.cliName,
        '--agent',
        Agent.claude.cliName,
        '--all',
      ]);

[Yusufihsangorgel]

Kept this consistent with the await subscription.cancel() in the sibling test above; switching both to addTearDown is a reasonable cleanup but out of scope for this PR.