Deadlock with backtrace-rs based stack unwinding

[sargarass]

I believe a deadlock can occur when one thread is making allocations while another thread is directly performing trace unwinding using backtrace-rs and also needs to allocate.

So, thread 1 acquires a lock with the following path:
step 1
step 2
step 3

And thread 2 acquires the lock in the opposite order.
step 1
step 2

It's interesting to note that using the backtrace-rs API with its inner lock is a bit weird because std also uses backtrace-rs for thread unwinding (link 1, link 2), but it employs a different lock for this purpose. Consequently, the lock inside backtrace-rs doesn't ensure exclusive access to its internals.