problem recumputing waveforms in a binary sorting analyzer

dear devs. when using a sorting analyzer in memory mode, i can correct spike times and recompute the waveforms.
unit_peak_shifts= score.get_template_extremum_channel_peak_shift(sorting_analyzer, peak_sign= 'neg')
sorting=spost.align_sorting(sorting, unit_peak_shifts)

sorting_analyzer.compute("waveforms", ms_before=1.5, ms_after=1.5, **job_kwargs)
sorting_analyzer.compute("templates", operators=["average", "median", "std"],**job_kwargs)

but in binary mode, after computing the waveforms in the first time, i cannot recompute them:

--> 294 sorting_analyzer.compute("waveforms",  ms_before=1.5, ms_after=1.5, **job_kwargs)
    295 sorting_analyzer.compute("templates", operators=["average", "median", "std"],**job_kwargs)
    296 print('computing template simliarity and metrics')
File ~\vspy\Lib\site-packages\spikeinterface\core\sortinganalyzer.py:1620, in SortingAnalyzer.compute(self, input, save, extension_params, verbose, **kwargs)
   1573 """
   1574 Compute one extension or several extensiosn.
   1575 Internally calls compute_one_extension() or compute_several_extensions() depending on the input type.
   (...)
   1617
   1618 """
   1619 if isinstance(input, str):
-> 1620     return self.compute_one_extension(extension_name=input, save=save, verbose=verbose, **kwargs)
   1621 elif isinstance(input, dict):
   1622     params_, job_kwargs = split_job_kwargs(kwargs)
File ~\vspy\Lib\site-packages\spikeinterface\core\sortinganalyzer.py:1697, in SortingAnalyzer.compute_one_extension(self, extension_name, save, verbose, **kwargs)
   1694     assert ok, f"Extension {extension_name} requires {dependency_name} to be computed first"
   1696 extension_instance = extension_class(self)
-> 1697 extension_instance.set_params(save=save, **params)
   1698 if extension_class.need_job_kwargs:
   1699     extension_instance.run(save=save, verbose=verbose, **job_kwargs)
File ~\vspy\Lib\site-packages\spikeinterface\core\sortinganalyzer.py:2654, in AnalyzerExtension.set_params(self, save, **params)
   2651 # this ensure data is also deleted and corresponds to params
   2652 # this also ensure the group is created
   2653 if save:
-> 2654     self._reset_extension_folder()
   2656 params = self._set_params(**params)
   2657 self.params = params
File ~\vspy\Lib\site-packages\spikeinterface\core\sortinganalyzer.py:2600, in AnalyzerExtension._reset_extension_folder(self)
   2598     extension_folder = self._get_binary_extension_folder()
   2599     if extension_folder.is_dir():
-> 2600         shutil.rmtree(extension_folder)
   2601     extension_folder.mkdir(exist_ok=False, parents=True)
   2603 elif self.format == "zarr":
File ~\.conda\envs\cupyenv\Lib\shutil.py:787, in rmtree(path, ignore_errors, onerror, dir_fd)
    785     # can't continue even if onerror hook returns
    786     return
--> 787 return _rmtree_unsafe(path, onerror)
File ~\.conda\envs\cupyenv\Lib\shutil.py:634, in _rmtree_unsafe(path, onerror)
    632             os.unlink(fullname)
    633         except OSError:
--> 634             onerror(os.unlink, fullname, sys.exc_info())
    635 try:
    636     os.rmdir(path)
File ~\.conda\envs\cupyenv\Lib\shutil.py:632, in _rmtree_unsafe(path, onerror)
    630 else:
    631     try:
--> 632         os.unlink(fullname)
    633     except OSError:
    634         onerror(os.unlink, fullname, sys.exc_info())
PermissionError: [WinError 32] The process cannot access the file because it is being used by another process: 'r\\[gpfs.corp.brain.mpg.de](http://gpfs.corp.brain.mpg.de/)\\stem\\data\\project_hierarchy\\data\\afm16924\\240526\\trial0\\postscript\\sorting_analyzer_results\\extensions\\waveforms\\waveforms.npy'

Start in spikeinterface/core/sortinganalyzer.py at SortingAnalyzer.compute_one_extension(), AnalyzerExtension.set_params(), and _reset_extension_folder(). Reproduce the binary-mode waveform recomputation on Windows using the reported sequence and inspect why waveforms.npy remains in use; done means recomputation completes without the PermissionError.