disable gate extension logic for the latest flowJo wsp

[mikejiang]

The existing gate extension rationale was detailed in RGLab/flowWorkspace#36, basically it is a heuristic fix for the issue of truncated gate coordinates recorded in flowJo xml workspace. i.e. whenever the negative gate coordinates observed from xml, it's an indication that the gate is meant to extend the far lower end to cover the rest of negative data area.

However, we started to experience some problems where the gate coordinate extension shouldn't be triggered.
e.g.
The exception will be thrown (RGLab/flowWorkspace#334) when attempting to extend ellipsoidGate, which appears to be unnecessary because ellipsoidGate is internally interpolated into polygon thus the coordinates are computed from antipodal points from xml and probably won't have the same truncation issue as the other polygonGates.

So we may want to skip extension for ellipsoidGate.

But the default gate extension behavior won't be even safe for polygonGate. e.g.
This is what the original gate looks like

And the gate coordinates do contain negative values

data.frame(gs_pop_get_gate(gs, "B8R+/Live/Q8: CD44- , CD62L-")[[1]]@boundaries, check.names = F)
  Comp-VioBlue-A Comp-BV510-A
1       45.16228     116.7371
2       45.16228      -1.3806
3       59.08485      -1.3806
4       80.12748     116.7371

And the default extend_value=0 threshold will trigger the extension logic, which distorted the original gate and its stats, e.g.

[mikejiang]

To investigate this behavior from the relative recent flowJo 10.6.1,

The gates seems to be fine without extension(by relaxing the extend_val threshold)

So we may conclude the legacy behavior has been fixed(i.e. gate original coordinates are always preserved by xml) by the newer flowJo so that we no longer need to do this hack.

For now, I will set default extend_val = -Inf to disable the logic. Once confirmed with @jspidlen, @jayalmarode about this , we can make this logic conditional on the flowJo wsp version.

[mikejiang]

Looks like it is still needed for the gate that is defined on log scale, even for the latest flowJo
e.g.

Without extension, gate includes less cells than it should

By extending the gate, the correct stats is restored

And we had the legacy flowJo xml that have the live/dead gate defined on biexp scale,e.g.

The coordinates are apparently wrong without extension.

So maybe for now, keep the current behavior as it is except skipping extending ellipsoidGate. When things go wrong, we will have to investigate the problem case by case and adjust extend_val as needed.