Fix threading bugs in workqueue.c causing ThreadSanitizer warnings

Title: Fix race conditions in workqueue.c causing ThreadSanitizer warnings

Description:

Problem

ThreadSanitizer detects two critical threading bugs in src/lib/evloop/workqueue.c:

1. Unlocking unowned mutex: Worker threads attempt to unlock control_lock during shutdown without having acquired it
2. Lock ordering inconsistency: Different lock acquisition orders between main thread and worker threads create deadlock potential

Impact

• Race conditions in production builds (ThreadSanitizer warnings indicate real bugs)
• Potential deadlocks during high worker thread turnover
• Undefined behavior when unlocking unowned mutexes

Reproduction

./configure --enable-gcc-hardening 'CFLAGS=-fsanitize=thread -g -O2 -fno-omit-frame-pointer' LDFLAGS=-fsanitize=thread --disable-asciidoc
./src/test/test_workqueue

Solution

Fixed both issues:

• Removed erroneous tor_mutex_release(&pool->control_lock) in worker exit path
• Reordered lock acquisition in threadpool_start_threads() for consistency

Patch

Note: I was unable to create a fork on this GitLab instance (possibly a configuration issue). As a workaround, the patch is available at: https://gitlab.com/hello12324/tor/-/tree/fix/workqueue-unlock-bug

If this causes any inconvenience, I apologize. I can also provide the patch via email or generate it with git format-patch if preferred.

Testing: All tests pass, zero ThreadSanitizer warnings (Note: test-memwipe fails but this appears to be a pre-existing issue with the test itself, which relies on undefined behavior by reading uninitialized and freed memory)

Environment

• Tor 0.4.9.3-alpha-dev
• Ubuntu 25.04, GCC 14.2.0