Fix gfp computation

[vferat]

As discussed in #160 and #195, introduces changes to the Global Field Power (GFP) computation based on the selected channel type.

When the channel type is eeg, the GFP is now computed using the standard deviation (std).
When the channel type is grador mag, the GFP is computed using the root mean square (rms).

[mscheltienne]

No need to nest here and you could use a dict to store the GFP functions per channel type:

_GFP_FUNC: dict[str, Callable] = {
    "eeg": lambda x: np.std(x, axis=0),
    "grad": lambda x: np.sqrt(np.mean(x**2, axis=0)),
    "mag": lambda x: np.sqrt(np.mean(x**2, axis=0)),
}

# and later to use it:
_check_value(ch_type, _GFP_FUNC, "ch_type")
gfp = _GFP_FUNC[ch_type](data)

It removes the additional private functions and the nesting of function within, and removes the if/else logic on channel types.

[mscheltienne]

This is good to merge? @vferat