Docs: a proven new test starts at tier B; tier C is for unproven tests

[lmoresi]

Tier C is virtually obsolete: it exists to keep an unproven test from derailing the code (decided 2026-10-11). The testing document's guideline "all new tests begin at tier C" contradicted that. It also contradicted the document's own C1/C2 sections, which call characterisations rare by design and give tier C its purpose: stopping a freshly written test, which may simply be wrong, from steering the implementation.

Changes:

• docs/developer/TESTING-RELIABILITY-SYSTEM.md: a new test that is proven starts at tier B and gates. Proven means a regression test written first and shown to fail without its fix, or a test asserting a hard baseline whose own correctness has been checked. Tier C stays for tests not yet proven.
• CLAUDE.md (Tests): the same rule, in two sentences.
• tests/test_0025_jit_compile_flags.py moves to tier B. It is the -fno-math-errno pin from JIT: compile kernels with -fno-math-errno by default (#834) #835, shown failing on development before its fix.

The style charter's tier table and its characterisation paragraph are unchanged; they do not say where new tests start.

Underworld development team with AI support from Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[lmoresi]

Review (docs and one test marker, before merge):

• test_0025 is proven. It failed on development before JIT: compile kernels with -fno-math-errno by default (#834) #835 for the right reason (the flag was missing), passed after it, and asserts the exact argument list. It now gates, since CI runs -m "not tier_c", and it still matches the tests/test_00[0-4]*py glob.
• Consistency: the guideline now agrees with the document's own C1/C2 sections and with the charter's tier table. Nothing else in docs/ tells a new test to start at tier C (searched for "Start at Tier C" and "begin as experimental").
• No library code changed.