fix(extensions): DataFilterExtension handling of filterSoftRange

[Pessimistress]

Closes #8774

Change List

• When filterSoftRange is truncated by filterRange fallback to use hard edge

[ibgreen]

A doc change warranted, or this is just an edge case?
Do we have a test harness?

[felixpalmer]

I wonder if it would be more efficient to keep the old shader and instead clip the value of the soft range in JS so it could never be in the wrong range

[Pessimistress]

@ibgreen no doc change required, this is a bug fix

@felixpalmer the issue here is when filter_min equals filter_softMin we still cannot use smoothstep.

[coveralls]

coverage: 90.004% (+0.002%) from 90.002%
when pulling 08f65c3 on x/data-filter-fix
into fa029dd on master.

[felixpalmer]

@Pessimistress I was suggesting we avoid filter_min === filter_softMin by clamping filter_softMin to filter_min + epsilon in JavaScript

[Pessimistress]

If you look at the original issue, they are using filter_min == filter_softMin == filter_softMax == filter_max. It would not work with any modification.
And that's not even getting into how to determine epsilon for an arbitrary float32...