[electrophysiology_browser] Dynamic channel metadata and visualization

[MaximeBICMTL]

Builds on #10403

Description

This PR adds support for usable MEG display in the electrophysiology browser by replacing hard-coded EEG values with dynamic values based on the channel metadata and signal.

Features

• Get the raw signal unit (V, T, T/m, s, ...) from the channel metadata instead of hard-coding V.
• Derive the display channel unit (µV, fT...) and multiplication factor from that unit.
• Compute the default channel amplitude from the signal values instead of using a hard-coded range for EEG.

Design choices

• The channel units are obtained form the BIDS channels.tsv fields, read through the LORIS API (just like the type in the channel types).
• The default channel amplitudes are computed per channel, not per channel type, this is because in the case of MEG, some channels can have largely different amplitudes than others.

Architecture

This PR also moves the channel metadata and BIDS channel information from the Redux-based state management to React contexts, which reduces prop drilling and makes them easier to use throughout the electrophysiology browser.

Backward-compatibility

This PR is backward-compatible with existing EEG acquisitions. If some metadata is not found, this PR falls back to assuming the acquisition is an EEG acquisition.

Performance:

This PR adds a O(n) scan on the signal values of visible channels to get the signal range. This did not cause noticeable lags in my tests even with 64 displayed channels.

Potential improvements

• I think it would be preferable to get all the channel metadata from the acquisition file rather than the BIDS metadata files, but doing so is non-trivial so I prefer to hold off for now.
• Maybe the electrophysiology browser should provide several channel amplitude strategies based on the user needs?
• Maybe use the extracted signal range for the whole channel rather than only the displayed range? This is more performant but results in significantly flatter curves when the display range is too small.
• Instead of absolute signal range, it might be preferable to use percentile-based signal range for the signal range in order to be more resistant to outliers, although I did not notice issues in my tests.

All these improvements can be added later IMO (although more metadata extract requires re-running the electrophysiology chunker script) and are easier to implement after this PR rather than before.

Screenshots

For the MPN MEG dataset:

[TODO]

For the EEG Face13 dataset:

[TODO]