Make msgp:sort directive more predictable when dealing with imports

Current implementation of `msgp:sort` directive used for canonical encoding and decoding by sorting map fields associate the types string names with it's SortType and LessFunction. This is done as commented in `elem.go` since map keys don't have references to the Elem associated with the directive.

Additionally msgp parses all import packages for the package it's generating methods which means that a `msgp:sort` directive exists for a given type already in the imported package it will be written first and then later overwritten by the main package being parsed which is always parsed last.

This can lead to interesting sorts of bugs where a `msgp:sort` can be missing in the parsed package but still be populated with potentially unexpected values from one of the imports.

We should guard against this either by not using sort directives from the import packages or by finding a way to make the process more robust and stop attaching the directives to strings representing type names.