temporary filename is non-random, causing bad state and the killthreads fedora flake #9

Bug

In this code:

If multiple requests run in parallel, they will both perform a file-system write to the same file. This sequence of events happens
in the fedora killthreads flake https://github.com/pwndbg/pwndbg/actions/runs/34255810405/job/102161276306 :

1. thread A starts a requests and starts writing to the .tmp.debuginfo file
2. thread A writes up to offset X
3. thread B starts a request and starts writing to the same .tmp.debuginfo file, truncating it in the process, and zeroing out everything that thread A wrote
4. thread A continues writing from offset X
5. thread B writes from the beginning of the file
6. thread A finishes, and GDB reads the DWARF while it has a zero-hole inside
7. symbol lookups don't work anymore
8. the expression pthread_kill() does not get evaluated due to a gdb error
9. our killthread function swallows up the error and doesn't think anything is wrong
10. but the thread doesn't actually get killed, and the assert fails

Repro

In the pwndbg root folder you can run:

docker compose run --rm -v $(pwd):/pwndbg fedora42 bash /pwndbg/repro-debuginfod-race.sh

with the repro file (slopped):
repro-debuginfod-race.sh

You'll see that debuginfod-zig repro's the bug while elfutils does not.

Fix

Use mktemp or equivalent.