Requested change by Tim Jarsky.

We want to continue the loop even if one single select dataset did not had
any epoch info.

Bug introduced in 595daef (MIES_SweepFormula_PSX.ipf: Add support for
multiple ranges, 2024-02-02).

We need to zap all null waves before storing the wave ref wave in the JSON
wavenote. The order does matter but SFH_GetSweepsForFormula also ignores
null ranges so we end up being ordered again in the end.

Bug introduced in 595daef (MIES_SweepFormula_PSX.ipf: Add support for
multiple ranges, 2024-02-02).

The additional /Z in the range based for loop is not required, but that is
a bug reported to WM.

Forgotten in 422eac8 (SF: Adapt SFH_EvaluateRange for multi dataset
input, 2024-01-20).

We are now removing empty entries in the output wave. This avoids an issue
with psxPrep which did not anticipate that.

As we are now not feeding in any null waves into PSX_OperationImpl we can
also remove the checks there.

The current implementation for the ranges did not handle multiple passed
epochs nor epoch wildcards. The check for intersecting ranges added in
6d10790 (PSX_OperationStatsImpl: Assert on intersecting ranges,
2024-02-06) was also broken.

While thinking about the current approach it was recognized that we
actually want to collect the event data from all sweep ranges of all
sweeps from the same equivalence class.

This has now been done. We also support list of epochs and epoch
wildcards.

This prepares for a future commit where we throw away single events and
thus PSX_AnalyzePeaks can result in no events event if we found peaks in
PSX_FindPeaks.

In a future commit we want to reuse these calculations for event
filtering.

We know how many entries in psxEvent and eventFit we have before starting
the loop so we can just call Redimension before and are done.

… amplitude

We need to filter that out first as we need the neighbouring distance to
the passing events.

And we also have to introduce an override setting as otherwise the tests
don't pass.

We only use the fixed value in case tau is NaN.

We can be called from PSX_PlotInteractionHook where we don't know if the
passed index is valid.

Let's do nothing if the index is out of range.

In dc9ed87 (SF_PreparePlotter: Don't kill and recreate the main panel
for SF_DM_SUBWINDOWS, 2024-06-19) we switched from killing the
sweepformula plot to clearing it.

But we forgot to remove additional controls and draw elements added by
e.g. PSX.

And now we also have to explicitly remove TraceUserData (TUD) as these are
not cleaned up automatically anymore, as killing an embedded subwindow
does not trigger the main window hook.

Due to the changed window handling in dc9ed87 (SF_PreparePlotter: Don't
kill and recreate the main panel for SF_DM_SUBWINDOWS, 2024-06-19) we need
to ensure that the host window of the SF panel is active.

This makes it easier to use and is more according to our philosopy that null
wave refs should be preferred over empty waves.

We only need index in the else branch and some more spaces are nice.

Using psxPrep requires sweep data from psx. But we used to only return
sweep data if we have found events.

We now always return sweep data so that psxPrep always works.

We now first offset and then filter. This requires that we also rename all
variables, constants and dimension labels.

Change requested by Tim Jarsky.