Generalizing suppression for pypocketfft.

[rwgk]

This valgrind error started appearing today, for completely unrelated changes. E.g.:

https://github.com/pybind/pybind11/pull/2895/checks?check_run_id=2061696834

==9513== 104 bytes in 1 blocks are definitely lost in loss record 6,250 of 8,539
==9513==    at 0x4A3852A: operator new(unsigned long) (vg_replace_malloc.c:342)
==9513==    by 0x153EFF33: PyInit_pypocketfft (in /home/runner/venv-3.9/lib/python3.9/site-packages/scipy/fft/_pocketfft/pypocketfft.cpython-39-x86_64-linux-gnu.so)
==9513==    by 0x4F5016: _PyImport_LoadDynamicModuleWithSpec (importdl.c:164)
==9513==    by 0x4F454A: _imp_create_dynamic_impl.constprop.0 (import.c:2297)
==9513==    by 0x4F44AD: _imp_create_dynamic (import.c.h:330)
==9513==    by 0x62546B: cfunction_vectorcall_FASTCALL (methodobject.c:426)
==9513==    by 0x4318A2: PyVectorcall_Call (call.c:231)
==9513==    by 0x431596: _PyObject_Call (call.c:266)
==9513==    by 0x4314E3: PyObject_Call (call.c:293)
==9513==    by 0x4D49C1: do_call_core (ceval.c:5092)
==9513==    by 0x4D32F7: _PyEval_EvalFrameDefault (ceval.c:3580)
==9513==    by 0x4C9161: _PyEval_EvalFrame (pycore_ceval.h:40)
==9513==    by 0x4C9161: _PyEval_EvalCode (ceval.c:4327)
==9513== 

[YannickJadoul]

For the record, we were already suppressing the same leak but with _imp_create_dynamic_impl.constprop.3 in the stack instead of _imp_create_dynamic_impl.constprop.0. This just generalizes that existing suppression rule to not care about that number. Hopefully that's better/enough to avoid future issues.

[rwgk]

Thanks a lot Yannick!