Catch any exceptions thrown during decoding data from disk

When any of the databases decode data from disk, they "graciously" handle failure (`fail`, `DeserialiseFailure`, `Left`, etc.). For example, the ImmutableDB and VolatileDB will truncate invalid blocks.

When a decoder throws an impure exception instead of a proper failure (`fail`), the exception is propagated instead of handled as "invalid data". When this happens while serving a request for a connection, the error policy will just restart the connection thread in question. However, when this happens on startup, e.g., during validation or recovery of the databases, the node will crash with that exception because the databases don't even get the chance to handle this invalid data. On the next startup(s), the same will happen unless the user intervenes.

Decoders should never throw impure exceptions, but bugs happen, see https://github.com/input-output-hk/cardano-base/pull/103.

We should catch such exceptions during decoding. Ideally this should be done in a central place, because it would be too easy to forget to do it in some decoder.