Boundary condition type updates

[timofeymukha]

This continues the series of PRs on cleaning up bc_t and descendants.

• A deferred destructor is added to bc_t and implemented in all types.
• The constructor and destructor in bc_t are called init_base and free_base to let init and base be deferred. This is similar to other hierarchies we have.
• The hierarchy is completely flattened, so everything descends from bc_t now. This is perhaps the easiest if we will have a common factory for all types. So a clean slate in that sense, and then we see if we can combine/remove some bcs or introduce and extra layer of inheritance.
• In symmetry, dong and non_normal we now have init instead of the individual constructor-like routine names. Hopefully to spread to all types under one interface.
• Similarly, the Dong bc uses JSON in the constructor as a preview for what will probably be the case for all types. Cleans up json parsing in fluid_scheme a bit.
• Rouge use of a local coef component was removed, using now the one in bc_t.

[timofeymukha]

A bit nitpick, but looks a bit strange that all bc's are calling init_base. Shouldn't the logic rather be init() in the type, and that init might call init() (can be call init_base) in the base type

So what I am aiming for eventually is that init is deferred in the base time and follows an abstract interface. Since you can't have a default implementation for a deferred routine, the name of the constructor for bc_t has to be called something else. We used init_base in a few hierarchies, so I picked it up here as well. So, then init will call init_base in the beginning to setup the base type. The deferred init is not quite there yet, but this PR is kind of prepping for that :-).

[timofeymukha]

@MartinKarp Whenever you have time :-). A new batch of updates is coming.