[VRP] Harden testcase safety check in long-lived bots

[PauloVLB]

Problem

Long-lived bots shouldn't be handling untrusted testcases locally (like during minimization, analyze, regression or variant tasks). The old check used the UNTRUSTED_UTASK env var with just a warning, which wasn't reliable.

Proposed Solution

Harden check_handling_testcase_safe in uworker_io.py to check testcase.trusted and environment.is_uworker(). If a long-lived bot (is_uworker() is False) tries to handle an untrusted testcase, it will immediately exit with a fatal error. Also aligned its behavior and return values with check_running_fuzzer_safe.

Notes on TODOs

• Removed the TODO in task_creation.py (create_variant_tasks_if_needed) because untrusted testcase handling is already covered.
• Updated the TODO in corpus_pruning_task.py to specify "engine fuzzing": the untrusted testcases we are supporting are strictly for blackbox fuzzing, which do not reach the corpus prunning task.

Testing

• Added unit tests in uworker_io_test.py covering trusted testcases, untrusted testcases on uworkers, and fatal exits on long-lived bots.
• Updated test helpers (test_utils.py, utask_helpers.py, minimize_task_test.py) to default testcases to trusted=True so unrelated tests don't trigger the fatal safety check.