Alter how quality metrics check extension dependencies

[chrishalcrow]

This PR changes how quality metrics are handled when an extension they depend on is missing.

The new behaviour is

• if you explicitly ask for a metric to be computed (si.compute_snr or analyzer.compute('quality_metrics', metric_names=['snr']) and don't have the extension computed, this will error.
• if you don't ask for any special metrics (analyzer.compute('quality_metrics')), this will compute all metrics it can. It will also gather all the metrics you can't compute, and why, and return a single warning.

The warning is e.g.

Some metrics you are trying to compute depend on other extensions:
    snr requires ['noise_levels', 'templates']
    amplitude_cutoff requires ['spike_amplitudes|waveforms', 'templates']
    amplitude_median requires ['spike_amplitudes|waveforms', 'templates']
    amplitude_cv requires ['spike_amplitudes|amplitude_scalings', 'templates']
    sd_ratio requires ['templates', 'spike_amplitudes']
    noise_cutoff requires ['spike_amplitudes']
To include these metrics, compute the required extensions using `sorting_analyzer.compute('extension_name')

Technically, we do this by doing a dependency check when we're figuring out which metrics to compute (done in _set_params of the ComputeQualityMetrics class). Then we don't include any metrics we don't have the extensions for.

If you call e.g. compute_snrs, it calls _has_required_extensions which does a check of the dependencies for a single metric, and raises the error message.

A nice consequence: now, the quality metrics extension has no dependencies, and will compute some spike-train related metrics if you give it an extension-less analyzer.

Made quite a big edit to the quality metrics modules docs page. Been meaning to do this for a while: https://spikeinterface--4080.org.readthedocs.build/en/4080/modules/qualitymetrics.html

We had discussed introducing a UnitMetric class or something. I think this is simpler for now!

[alejoe91]

That's good for me! For the spike_amplitudes/amplitude_scalings maybe we can do required_extensions=["spike_amplitudes|amplitude_scalings"] like we have in the core sorting analyzer here?