CI: Fix problems with valgrind memory leak check

Two problems:

• When moving to pybind11, we renamed our binding's .so file, so the leak checker incorrectly ignores all PyTango leaks. This name check has been fixed.
• Under Python 3.14 multiprocessing uses forkserver instead of fork, and this causes valgrind to exit before the test code is finished running. Then we cannot parse the incomplete XML files. The script now waits properly before reading the XML files.

A few additional tests were added for things that could leak.

For the linux:test-main-cpptango job we now include the debug symbols so that the leak checker output is more useful.

TODO:

• add a debug method to force a leak, and verify that our leak checker finds it (prevent regressions)