Double-check `cfg_version` test coverage

[jieyouxu]

          Discussion: this test is kinda testing multiple things and bears multiple responsibilities, we may want to split this test up:

1. We wish to exercise the feature-gating of the unstable feature itself. That is, I would expect #[cfg(version("1.49.0"))] to be this feature-gate test without #![feature(cfg_version)] and the user should get the "this is unstable" error message.
2. We wish to exercise the wellformed-ness of the values passed to version(..) as a sort of "interface". In particular, I'd consider splitting tests for:
   1. The syntactical wellformedness:
      • Proper name-with-paren-value form (version(..)): known vs unknown major version string "1", known vs unknown major-minor version string, known vs unknown major-minor-patch version string, empty version string, some "hello world" nonsense version string.
      • Invalid forms (version = ".." or bare version).
   2. The semantic wellformedness of the version string: major version (e.g. "1"), major-minor version (e.g. "1.1"), major-minor-patch (e.g. "1.1.1"), and also known-vs-unknown variants for them.
3. I'd also check that if user manually specifies --cfg version or --cfg version=".." they are not affected and usable with or without feature gate (I need to double-check that).

This is also fine as a follow-up, I can probably write those coverage.

Originally posted by @jieyouxu in #141413 (comment)