tests: run_all_tests.sh invokes test_terminal.eigs unguarded — wedges forever when stdin is a held-open pipe

[InauguralPhysicist]

Found by @Nitjsefnie while working on #653 (see their report) — they hit it live as an 11+ minute hang.

tests/run_all_tests.sh:2073 is the last unguarded interpreter invocation in the runner:

TM_OUTPUT=$(./eigenscript ../tests/test_terminal.eigs 2>&1); TM_OUTPUT_RC=$?

Compare the guarded shape at :93:

out=$($EIGS_TMO ./eigenscript "../tests/$file" </dev/null 2>&1); rc=$?

Two things missing: $EIGS_TMO (the runaway backstop from #616) and </dev/null. test_terminal.eigs reads stdin, so in any environment where stdin is a held-open pipe — a captured or backgrounded CI-ish shell — it blocks forever. No timeout wrapper means the whole suite wedges rather than failing.

We have been treating this as folklore instead of fixing it. CLAUDE.md carries the workaround as a standing rule:

make test must run with stdin available or redirected from /dev/null — test_terminal.eigs blocks forever reading a pipe that never EOFs.

That is a bug documented as a constraint. The runner should be robust on its own rather than requiring every caller to remember an invocation discipline.

Fix

Give :2073 the same treatment as :93 — $EIGS_TMO and </dev/null. Then delete the CLAUDE.md rule, since it stops being true.

Done when

• :2073 uses $EIGS_TMO and </dev/null
• [60] Terminal Builtins still reports 10/10 in a normal run
• The suite completes (does not hang) when run with stdin held open — e.g. bash -c "sleep 300 | bash run_all_tests.sh", which wedges before the fix
• The CLAUDE.md stdin rule is removed in the same PR

Notes

Scoped to one line plus a doc deletion; the verification above is the interesting part. @Nitjsefnie has first refusal since they found it.