Throw from readDeck() instead of std::exit()

[hakonhagland]

Alternative to #6933.

Problem

When readDeck() encounters unrecoverable parse errors, it calls MPI_Finalize() + std::exit(EXIT_FAILURE). This is problematic because:

• Unit tests fail silently — std::exit() bypasses Boost.Test, producing exit code 1 with no test failure report
• Not idiomatic C++ — functions should signal errors via exceptions, letting callers decide how to handle them
• MPI cleanup is wrong — MPI_Finalize() requires cooperative shutdown from all ranks; on an error path MPI_Abort() is more appropriate

Solution

• readDeck.cpp: Replace MPI_Finalize() + std::exit(EXIT_FAILURE) with throw std::runtime_error(failureMessage)
• Main.hpp: Widen catch in Main::initialize_() from std::invalid_argument to std::exception so it catches both existing and new exception types. The existing MPI_Abort() call in the catch block handles MPI cleanup correctly.

Impact

Context	Before	After
Production	MPI_Finalize + std::exit(1)	Exception caught by Main::initialize_(), MPI_Abort, return false
Unit tests	Silent exit code 1	Boost.Test catches std::runtime_error, reports failure with message
Normal runs	No change	No change (error path only)

[hakonhagland]

jenkins build this please

[bska]

I still maintain that the unit test justification is just wrong. You should never add a unit test that has parse errors.

[hakonhagland]

You should never add a unit test that has parse errors.

Yes, but in my opinion during development it will be helpful to have this to catch deck errors quickly.

[bska]

You should never add a unit test that has parse errors.

Yes, but in my opinion during development it will be helpful to have this to catch deck errors quickly.

And you're making everything more difficult for the slight convenience of something that happens very infrequently. You should use other tools when developing.