Add concatenated to _get_data for amps and locs; use in _get_amplitudes_by_units

[chrishalcrow]

Closes #4028

PR implements a concatenated argument in get_data method for the spike_ampltiudes and spike_locations extensions. This returns the data per unit, concatenated per segment. I think this is a helpful thing to have, and was easy to implement.

Then used this new functionality in _get_amplitudes_by_units.

Benchmarked using:

import spikeinterface.full as si
from spikeinterface.qualitymetrics.misc_metrics import _get_amplitudes_by_units
from time import perf_counter
import numpy as np

# an hour-long NP2-kilosort4 analyzer
sa_path = "/Users/christopherhalcrow/Downloads/kilosort4_sa"

times = []
for _ in range(10):
    
    sa = si.load_sorting_analyzer(sa_path, load_extensions=False)
    sa.load_extension("spike_amplitudes")
    t1 = perf_counter()
    amps = _get_amplitudes_by_units(sa, unit_ids=sa.unit_ids, peak_sign="neg")
    t2 = perf_counter()
    times.append(t2-t1)
    del sa

print(np.median(times))

On main: 1.68s
This PR, without numba: 1.13s
This PR, with numba: 0.0806s

So a 20x speedup if numba is around :)

[samuelgarcia]

Merci Chris. This is perfect.
@alejoe91 : I let you have a look.