faster validation

[odanoburu]

ideally the validation should be fast enough for interactive use (e.g., in the emacs mode), but is currently somewhat slow (good enough for validating in continuous integration, though)

[odanoburu]

#29 implements a cache for the partial indices of each lexicographer file; when validating, files that haven't changed are not re-parsed, but have their partial indices read from disk, merged, and used for validation.

profiling shows that the main bottleneck is now in I/O (reading the cache). using a faster serialization library will help improve performance, but maybe there's a better approach.

-- edit: we may use (phony?) targets to rebuild in-memory haskell values; this + IPC will allow us to forgo I/O. we would need to add a dependency for IPC, though (probably an HTTP server)