Restore Check 0.9.4 compatibility with a ck_assert shim in the test header

[zandbelt]

Follow-up to #159 and #161, which fixed #91 by raising the documented Check minimum from 0.9.4 to 0.12.0. This restores the 0.9.4 minimum for the 0.x line without touching the test code: the tests never actually built against 0.9.4, because the ck_assert family (ck_assert, ck_assert_msg, ck_assert_int_eq, ck_assert_str_eq) only arrived in Check 0.9.6, and ck_assert_uint_eq in 0.9.10.

Two commits:

1. test: provide the ck_assert family for Check 0.9.4 to 0.9.9. Twenty-three lines in check_cjose.h under #ifndef: ck_assert_msg, ck_assert, ck_assert_int_eq, ck_assert_uint_eq and ck_assert_str_eq, using the definitions of the release that introduced each (the integer comparisons print intmax_t / uintmax_t like later releases do). The .c test files are unchanged.
2. Revert the 0.12.0 bump in README.md, CMakeLists.txt, configure.ac and configure, with the rationale in the commit message. RHEL 6 ships Check 0.9.8, RHEL 7 0.9.9, RHEL 8 0.12.0 (checked against the CentOS vault and Rocky mirrors); 1.0.x remains free to raise the floor.

Changes since the first revision, per the review: ck_assert_uint_eq joined the shim so the two uses in check_concatkdf.c stay as they are (the #159 cherry-pick is gone), and the separate Check CI workflow is dropped; the regular CI keeps compiling and running the tests against the distro's Check.

Verified locally against Check 0.9.4, 0.9.6, 0.9.8, 0.9.9, 0.9.10, 0.9.14, 0.10.0, 0.12.0 and 0.15.2: all pass. Without commit 1, 0.9.4 fails to build on ck_assert_str_eq and 0.9.6 to 0.9.9 on ck_assert_uint_eq; with it, a deliberately wrong expected length reports Assertion 'otherinfoLen==24' failed: otherinfoLen==23, 24==24 on 0.9.4 and 0.9.9 exactly as 0.9.10 does.

🤖 Generated with Claude Code

[zandbelt]

as mentioned elsewhere, this would be helpful for 0.7.x to be a complete drop in (upgrade) for 0.6.2.x

[TheStormN]

Ok, I will have a second thought about it, but isn't it a separate pipeline just for Check a bit overkill? While I understand the need for OpenSSL, having the unit tests compiled and run during standard CI seems good enough to me.

[zandbelt]

Agreed, the Check workflow is dropped; the PR is now just the shim in check_cjose.h plus the revert of the 0.12.0 bump, rebased on current master. The regular CI compiles and runs the tests against the distro's Check as before.