Add correct_inter_session_displacementfunction

[JoeZiminski]

This PR requires #3231 , on which it is rebased.

This draft PR starts to play around with a rough implementation of 'inter-session displacement' estimation / correction, as discussed #2626.

The problem to be solved is inter-session movement ('displacement') of the probe. This is very similar problem to the inter-session motion correction, and can use many of the same tools. The idea is accept as input a list of recording objects (and, optionally, a list of intra-session motion correction outputs). For each session, an activity histogram will be generated, and then the inter-session drift and necessary correction estimated. The correction is then applied to the recording object, but a displacement_info object (similar to motion_info object) will be output. This could be used to correct peaks directly and optionally later, correct templates (see #2626).

At first I'll just do a very straightforward, rough implementation, then work on improving it. The steps are:

1. Accept a list of recordings and optionally a list of associated motion_info dicts (output from intra-session motion correction). If motion_info is not passed, peaks and peak locations will be estimated. (to improve: here just use the entire reocrding, but in future can use subsets of the recording, in a way that maximises the accuracy of the estimation).
2. With the peaks and peak locations, generate 2d or 3d activity histogram for the session (again, the entire session is used here, can be improved).
3. Run an optimisation to estimate displacement and correction between sessions from the histograms. This should limit the amount of data lost at the border as well as move each session a similar amount (so interpolation artefacts are comparable same across sessions)
4. If motion_info is provided, update the drift estimates with the inter-session drift and return. Otherwise, make the recordings InterpolateMotionRecording and return. Optionally return the displacement_info which copies motion_info.

This is currently very-rough draft PR but of course welcome feedback!

TODO:

• TODO: DREDGE on concatenated data is working very well.
• option to restrict 'interpolation' to channel multiple (i.e. no actual interpolation, just shifting channels)
• do a simple argmax over non-rigid with say 2-20 probe segments and see what works best (smoothing too)
• interpolate templates

  spikeinterface/src/spikeinterface/generation/drift_tools.py

  Line 4 in 0e13973

  from typing import Optional

FUTURE:

• more features as Nick Watters mentioned, possibly discuss adding this to machinery time permitting

[JoeZiminski]

missing "entire_session"

[JoeZiminski]

delete this

[JoeZiminski]

this is wrong now