Raise SamplingError if model doesn't have free RVs

[stestoll]

This small PR changes sample() such that it checks for free RVs in the model immediately (before checking start point), and raises a SamplingError instead of a ValueError if the model doesn't have any free RVs.

I think raising a SamplingError is better than a ValueError, since nothing value-wise is wrong with the model itself, just sampling is not possible. Also, SamplingError is more specific than ValueError.

[codecov]

Codecov Report

Merging #4955 (09d5d73) into main (4b57c2f) will decrease coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main    #4955      +/-   ##
==========================================
- Coverage   74.08%   74.06%   -0.03%     
==========================================
  Files          86       86              
  Lines       13862    13862              
==========================================
- Hits        10270    10267       -3     
- Misses       3592     3595       +3     

Impacted Files	Coverage Δ
pymc3/sampling.py	85.67% <100.00%> (ø)
pymc3/backends/report.py	89.51% <0.00%> (-2.10%)	⬇️

[michaelosthege]

Looks good. The message is also easier to understand.

[canyon289]

LGTM