C-backend tests still hand cc the clang-only flag: #4832 added get_cc_args() and no test calls it (test-ratchet red on master, 25 tests) #5907

Part of #5906 (row 1a). Re-opens the half of #4832 that its closing commit did not do.

Symptom

Bootstrap Test Ratchet (.github/workflows/bootstrap-tests.yml, job test-ratchet) is red on master: the failing set grew by 25. Master run 37066722367 (1cf5252dc, 2026-10-02) and PR #5902's run list the same 25 tests (the two sorted lists are identical). Each panics with:

cc: error: unrecognized command-line option '-ferror-limit=0'

The last green master run of this workflow is 2026-09-07.

The 25: tests/c_auto_type_array_literal.rs (3), c_enum_member_access.rs (3), c_enum_path_members.rs (1), c_literal_list_type.rs (8), c_module_array_consts.rs (4), c_static_array_params.rs (1), c_tuple_positions.rs (3), corpus_unresolved.rs (1), icarus_lowerable.rs (1).

Cause

#4832 diagnosed it: -ferror-limit= is a clang flag and CI's cc is gcc. It was closed on 2026-09-26 by 1acab8481, which changed two files: it added bootstrap/tests/common/mod.rs::get_cc_args() (probes cc for -ferror-limit=0, then -fmax-errors=0) and a docs/now note. No test was switched to it:

git grep -l 'common::' origin/master -- 'bootstrap/tests/c_*.rs'      -> 0 files
git grep -l 'ferror-limit=0' origin/master -- 'bootstrap/tests/*.rs'  -> 16 files

#4832's second half therefore still holds: a test that counts file:line:col: error: lines passes on gcc without compiling anything, because gcc's refusal of the flag has no line:col.

Fix plan

Every bootstrap/tests file that runs cc takes its arguments from common::get_cc_args(). Reproduce on this Mac with gcc as cc (a PATH shim to gcc-15), before and after. A negative control: a C snippet with a real error must fail the count-shaped tests under gcc after the fix (it passes them today).

Acceptance

• git grep -n 'ferror-limit' -- bootstrap/tests names only common/mod.rs
• with gcc as cc: the 25 pass, or each one still failing is a named C-backend defect with its own issue and a baseline row that says why
• the negative control fails before the fix's helper is used and passes after (shown in the PR)
• test-ratchet green on the PR