Unify Error Handling

The parser & typechecker currently make use of the `Err` monad from `ErrM`. The analysis backends currently have no consistent error handling approach and mostly just call `error` if something unexpected happens.

We should generalise the `Err` monad so that it can be used by the various backends and rework them so that functions that can fail return an `Err`.

Start by reading the existing Err monad in ErrM and how the parser and typechecker use it. Then inspect the analysis backends to identify functions that currently call error; done means the backends consistently return Err for operations that can fail.

Written by the indexing model from the issue text.