fix(BaseModel): don't suppress error if exe not found

[wpbonelli]

close #1896

[codecov]

Codecov Report

Merging #1901 (8e0ad42) into develop (ca838b1) will increase coverage by 1.7%.
The diff coverage is 100.0%.

@@            Coverage Diff            @@
##           develop   #1901     +/-   ##
=========================================
+ Coverage     70.8%   72.6%   +1.7%     
=========================================
  Files          257     257             
  Lines        56216   56216             
=========================================
+ Hits         39810   40817   +1007     
+ Misses       16406   15399   -1007     

Files Changed	Coverage Δ
flopy/mbase.py	69.4% <100.0%> (+0.1%)	⬆️

... and 55 files with indirect coverage changes

[wpbonelli]

MF6 autotests are failing here because some of them try to initialize Modflow models without specifying an exe. Previously failures to resolve the default exe mf2005 were suppressed on model init/load.

I see two options:

1. Require a valid exe at init and/or load time. If we go with this, this PR is ready as is, but mf6 autotests need a fix.
2. Defer exe resolution until run_model(). It may be nice to allow loading without specifying an exe, e.g. just to load a model, edit some settings, and rewrite input files. This PR will need changes, but mf6 autotests should not.

[langevin-usgs]

@w-bonelli, what if we issue a warning at init/load time if the exe is not available, and issue an error in run_model if the exe is not available? The only potential problem is that there might be cases where the exe name is used to determine how to load a package or what options are supported.

[wpbonelli]

Updated to warn on init/load, and error on run.