Would close #4261

This PR allows you to remove spikes from a unit during curation by specifying the following in your curation json file:

"discard_spikes": [
    {
        "unit_id": "u10",
        "indices": [
            56,
            57,
            59,
            60
        ]
    },
    {
        "unit_id": "u14",
        "indices": [
            123,
            321
        ]
    }
]

This new feature means that the curation format gets bumped to v3!

You can discard at the same time as merging and splitting.

Tricky bit that took me ages to think through: how to apply it to an analyzer. Decided to discard spikes at the same time as splitting. During the splitting step, we re-wrangle discard spikes into another split unit (call them "discard units") and keep track of the discard unit id. Then remove the full discard units after the splitting. This allows us to use the existing splitting machinery (including splitting extension etc) for discards - nice!

We have to do merges after splitting+discarding. This is because the spike indices change after merging. To avoid wrangling spike indices (gross!) we just do discarding first.

Tests to do:

• test curation format
• test v2 -> v3 update
• test actual discarding when also splitting / merging