Add option to flatten indexes when creating epochs with `event_id`

[cbrnr]

When creating epochs from a subset of the given events (using the event_id argument), the epoch indexes correspond to the indexes from the original events array. Here's an example:

import numpy as np
from mne import Epochs, create_info
from mne.io import RawArray
from numpy.random import default_rng

n_channels = 3
duration = 25
sfreq = 250
data = default_rng(42).standard_normal(size=(n_channels, duration * sfreq)) * 5e-6
info = create_info(n_channels, sfreq, "eeg")
raw = RawArray(data, info)

n_events = 9
events = np.column_stack(
    (
        np.arange(1, n_events + 1) * sfreq,
        np.zeros(n_events, dtype=int),
        np.tile([1, 2, 3], n_events // 3),
    )
)

epochs = Epochs(raw, events, event_id=2, tmin=-0.2, tmax=0.5)
epochs.plot()

This results in three epochs with indexes [1, 4, 7], for example as shown in the x-axis labels of the epochs plot:

I think this behavior is not very intuitive, as users will probably expect epoch indexes to range from 0 to N - 1, i.e., [0, 1, 2] in this example.

My current workaround is to not use event_id at all, but pass a subset of the original events array. Although this works, it feels like there should be an easier option to achieve this behavior with event_id. For example, we could add a new parameter flatten_indexes=False (feel free to suggest a better name). WDYT?

[larsoner]

I could see use cases where users might want what we currently do, and others might want them renumbered ascending.

Instead of a new constructor option (which already has too many options!) how about a epochs.reset_indices() or similar? Maybe this will need to reset stuff like drop_log etc., not sure...

[cbrnr]

Instead of a new constructor option (which already has too many options!) how about a epochs.reset_indices() or similar? Maybe this will need to reset stuff like drop_log etc., not sure...

Yes, this would work of course! I can take a stab and see what else needs to be changed.