Replace LineStack with MultiLineGraphic for performance

[cboulay]

After migrating phosphor from custom wgpu shaders to fastplotlib, performance regressed for dense multichannel displays. The root cause is that LineStack creates separate LineGraphic objects per channel, and the update loop in _update_graphics() performs N separate GPU buffer assignments:

for ch in range(buf.n_visible):
    self._line_stack[ch].data[idx_start:idx_end] = data_slice[ch]

@pperanich's MultiLineGraphic in his fastplotlib fork packs all lines into a single GPU buffer, allowing bulk updates in one assignment. This plan adopts his fork as an immediate performance fix.

See ezmsg-org/ezmsg#229