Make ml_progLib run type inferencer on each call

It would be nice to have `process_topdecs` fail immediately when given type-incorrect code to parse. This would be particularly helpful when developing code in `basis`, but is not generally desirable in the bootstrap translation.

I suspect this feature could be implemented by adjusting `ml_progLib` to keep track of the type inferencer state produced by the declarations that have been accumulated this far.

There needs to be a way to turn this feature off, e.g. by making the the type inferencer state information optional: `NONE` means that type inferencer is not run.

Start by locating ml_progLib and its process_topdecs entry point, then read how accumulated declarations and type-inference state are represented. Determine how an optional inferencer state would be threaded through calls and how bootstrap translation remains unaffected. Done means type-incorrect input can fail immediately when enabled, while the feature can be disabled.

Written by the indexing model from the issue text.