fix: enhance version input handling

[shenxianpeng]

The following review suggestion comes from Codex

---

• The new installer flow introduces at least two functional regressions: it rejects valid
semantic version inputs and it can fail after a successful package-manager install
because fallback installers are invoked unconditionally. Those cases make the composite
action less reliable than the base branch.

Full review comments:

• [P2] Preserve semantic version strings for tool installation — /Users/sxp/Repos/cpp-
  linter/cpp-linter-action/action.yml:384-390
  This now hard-fails any version like 20.1 or 20.1.7 unless it happens to be an existing
  path. That is a regression from the previous clang-tools -i ${{ inputs.version }}
  behavior: both the clang-tools and clang-tools-wheel CLIs accept semantic version
  specifications, and cpp-linter itself resolves dotted versions when selecting clang-
  format/clang-tidy. Workflows that pin an exact wheel/build version will now stop here
  with Invalid version input instead of installing the requested tool.
• [P2] Stop after the first successful tool source — /Users/sxp/Repos/cpp-linter/cpp-
  linter-action/action.yml:412-427
  On Linux and macOS this loop runs even when the earlier apt/Homebrew step has already
  installed the requested clang tool. As a result, a later PyPI/static install failure
  will now fail the action even though clang-format-$version/clang-tidy-$version is
  already present, which breaks the documented “try sources in order” fallback behavior.
  The secondary installers should only run when the prior source did not yield a usable
  tool.

[2bndy5]

The description here is typical AI slop. The unnecessary wordiness... The reasoning shown doesn't demonstrate an understanding of the code it is modifying.

Some of the changes here make a little sense. But, they are lazy changes that hardly achieve the stated goals.

[2bndy5]

This is ok, but it isn't complete semantic version support.

Because it only checks the major version is an int, users could pass in version: 12.x, and none of our install tools support this format.

[shenxianpeng]

thanks for taking time to review 😃 so let's just close it.

[2bndy5]

There's nothing really wrong with #368. I've tested it in the test repo and verified it should be backward compatible. Maybe you want to play around with test-wheels branch on the test repo? Otherwise, I'm confident enough and ready to move #368 forward.

FYI, cpp-linter/cpp-linter-rs#279 aims to remove all this condition logic (and version parsing/handling) from action.yml. Instead, it would be all done in rust; no pip/Python or nushell really needed. At least that's the goal.

[shenxianpeng]

Thank you for your testing. let's move forward. And looks forward to have cpp-linter/cpp-linter-rs#279 to remove all the condition logic in the future.