Having this catch-all made it difficult to develop and troubleshoot
nixops.

For example, when adding a new backend and forgetting to add the
corresponding `option` in the `.nix` file, nixops would die simply
with

    error: 'serverType'

which is not a good indication over what went wrong.
Specifically so because `error(str(e))` removed the type of the
exception, turning `KeyError: 'serverType'` into just `serverType`.

So far, you could pass `--show-trace` or `--debug` to see
Python stack traces.
However, that requires the error to be easily reproducible
(when you see it, you have to run again with `--show-trace`),
making debugging of rare problems hard.

For an ops tool we should make debugging of rare/unknown
problems as easy as possible.
Consequently, this commit makes Python raise the `Exception`
with stack trace if we have no idea at all what the exception
type is.

The other cases (when we know what exception type it is)
remain unaffected.

nixops script: Remove top-level Exception catch-all.