Update to Doctest 2.5.3

[nlohmann]

Also try to remove flag exclusions.

[nlohmann]

Closing unmerged — hit a hard upstream blocker.

Full-matrix CI surfaced two real regressions from the doctest 2.5.3 bump itself (not from the CI-flag cleanup in this PR, which was independently verified safe):

1. GCC 4.8/4.9 permanently broken. doctest's filldata<T> stringification hits an ambiguous template instantiation on GCC 4.8/4.9 (confirmed via our unit-bjdata failure on std::vector<unsigned char>). This is doctest#1112, "introduced in 2.5.0, and is present in all subsequent versions" — and doctest maintainer mitchgrout confirmed GCC 4.x support was intentionally and permanently dropped in doctest#1075, not a bug they intend to fix. Our ci_test_compilers_gcc_old matrix tests 4.8/4.9/5/6 to back the README's claimed GCC 4.8–14.2 support range.

2. Clang 22 (dev) -Wshadow failure, lower priority: doctest #1079 deliberately swaps DOCTEST_COUNTER from __COUNTER__ to __LINE__ on Clang ≥22, breaking anonymous-symbol uniqueness for our SETUP_TESTCASES() macro pattern. silkeh/clang:dev is an unpinned nightly, so treating this as toolchain churn rather than blocking.

Any future doctest 2.5.x bump needs an explicit decision on dropping GCC 4.8/4.9 from officially-tested support first — that's a project support-policy call, not something to fold into a test-framework version bump. Tracked in the project todo list.

🤖 Generated with Claude Code