BUG: simulation smoothed measurement disturbance with FILTER_COLLAPSED

[ChadFulton]

Closes #4800.

The problem here was that FILTER_COLLAPSED makes the effective number of endogenous variables smaller than the actual size of the observation array, but the simulation smoother wasn't taking this into account. This PR:

• Makes the simulation smoother behave appropriately with FILTER_COLLAPSED
• Adds a test for the smoothed measurement disturbance and the simulation smoothed measurement disturbance.

[ChadFulton]

Many thanks to @bashtage for pointing the bug out.

[coveralls]

Coverage decreased (-0.06%) to 83.59% when pulling bf9f9fe on ChadFulton:gh-4800 into f6a2181 on statsmodels:master.

[codecov-io]

Codecov Report

Merging #4810 into master will increase coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #4810      +/-   ##
==========================================
+ Coverage   81.13%   81.14%   +<.01%     
==========================================
  Files         581      581              
  Lines       89874    89895      +21     
  Branches    10040    10041       +1     
==========================================
+ Hits        72919    72944      +25     
+ Misses      14676    14674       -2     
+ Partials     2279     2277       -2

Impacted Files	Coverage Δ
statsmodels/tsa/statespace/simulation_smoother.py	90% <ø> (ø)	⬆️
statsmodels/tsa/statespace/tests/test_collapsed.py	96.55% <100%> (+1.58%)	⬆️
statsmodels/stats/descriptivestats.py	24.13% <0%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f6a2181...bf9f9fe. Read the comment docs.

[ChadFulton]

Green except for coveralls, which I think is incorrect (I don't really understand coveralls output most of the time)