[ENH] Improve _get_series_len performance

[HippocampusGirl]

Fixes #2405

Changes proposed in this pull request

When dummy_scans is provided in the configuration, we do not need to estimate the number of non-steady-state volumes, thus increasing performance of workflow creation.

This effectively halves the time needed for init_func_preproc_wf in my test dataset.

Documentation that should be reviewed

Not applicable

[effigies]

The config object is a hulking beast, and I'd prefer to keep its tentacles confined to init_*_wf() functions, and pass arguments to standalone functions, to make them easier to reason about.

def _get_series_len(bold_fname, skip_vols):
    from niworkflows.interfaces.registration import _get_vols_to_discard
    img = nb.load(bold_fname)
    if len(img.shape) < 4:
        return 1

    if skip_vols is None:
        skip_vols = _get_vols_to_discard(img)

    return img.shape[3] - skip_vols

And then, above, pass config.workflow.dummy_scans to _get_series_len().

[oesteban]

We probably want to make an overhaul of the documentation, on the code guidelines section) and:

• mention this (please do not use config deep in the workflow)
• recommend black for everything except the workflow.connect function

[HippocampusGirl]

I haven't actually looked at the code guidelines in a while, so I read them again just now.

I realized that I probably also should have considered the branch naming/pull request naming, and added [ENH].

[effigies]

This is great! I appreciate the attention you pay to details like this that improve performance.

[HippocampusGirl]

Thank you :-) I actually only noticed this by accident while profiling something else.