SIGUSR1 leak-only sanitizer output is classified as a hard failure

[Nitjsefnie]

Describe the bug
On current main, the multithreaded SIGUSR1 sanitizer subtest records normal ASan-integrated leak-only output as FAIL: sigusr1-mt: ASan/UBSan report on stderr instead of accepting it as the tolerated leak case. The same output contains both ERROR: LeakSanitizer: detected memory leaks and the standard SUMMARY: AddressSanitizer: line, and the tolerated result is never observed for that report shape.

This is distinct from #945/#953, which covered tests/test_doc_examples.py. It was discovered during an independent review of unfiled sanitizer test-harness findings on 2026-08-18.

To reproduce
From the repository root, replay the current predicates with representative ASan+LSan output:

err=$(mktemp)
printf '%s\n' \
  'ERROR: LeakSanitizer: detected memory leaks' \
  'SUMMARY: AddressSanitizer: 16 byte(s) leaked in 1 allocation(s).' >"$err"

if grep -qE 'AddressSanitizer|runtime error:' "$err"; then
  echo 'hard failure'
elif grep -q 'LeakSanitizer: detected memory leaks' "$err"; then
  echo 'tolerated leak'
fi

The command prints hard failure.

Expected behavior
Leak-only output from the repository's ASan build is accepted by the explicit LeakSanitizer tolerance path. Actual ASan or UBSan errors still fail the subtest.

Actual behavior
The leak-only report is classified as an ASan/UBSan failure before the LeakSanitizer tolerance result can be selected.

Environment

• OS: Linux; the classification is shell/grep control flow and is not architecture-specific
• GCC version: not applicable; the minimal reproduction does not compile the project
• EigenScript version: main at a916975b765581b6c531174ba28a67a05d5c277f